How to Measure Whether AI Search Is Recommending Your Business

How to Measure Whether AI Search Is Recommending Your Business

A business owner in Glendale searches Google for their company.

The website appears. Their Google Business Profile looks complete. Service pages are indexed, and Google Search Console shows impressions.

Everything seems reasonably healthy.

Then they open ChatGPT and ask:

Can you recommend a reliable website design company in Glendale, California?

Several businesses appear in the answer.

Their company isn't one of them.

That raises a question traditional SEO reports don't fully answer:

When people ask AI to recommend a business, how do you know whether your company is actually being suggested?

And if you make improvements to your website, how do you tell whether anything changed?

The answer requires a different kind of measurement.

Traditional search reporting is still important. Rankings, impressions, clicks, local visibility, and conversions remain useful.

But AI search visibility introduces new questions:

• Does the AI know your business exists?

• Does it accurately describe your services?

• Does it cite your website?

• Does it actively recommend you?

• Which competitors appear instead?

• Are customers reaching you through AI-generated answers?

You can measure parts of this today.

Other parts remain difficult to observe because businesses do not have unrestricted access to every AI-generated conversation or recommendation.

The practical approach is to combine controlled testing, official platform reporting, website analytics, and customer acquisition data.

What Does It Mean When AI Recommends Your Business?

The first mistake to avoid is treating every appearance of your business name as a recommendation.

Consider these examples.

Brand mention

An AI response says:

Navasartov is a company providing web development and custom software services.

That demonstrates some recognition of the business.

It does not mean the AI chose the company over competitors.

Website citation

An AI answer about PHP development links to a Navasartov technical article.

The article has been used or presented as a supporting source.

That is citation visibility, not necessarily commercial endorsement.

Direct recommendation

Someone asks:

Which companies should I contact to build a custom customer portal?

The AI includes your business among relevant providers to consider.

That is a business recommendation.

AI referral

Someone clicks a link from an AI answer and visits your website.

That is a measurable website acquisition event when the referral information is preserved.

AI-influenced customer

Someone learns about the business through AI, later searches its name on Google, and submits a contact form.

AI may have influenced the decision, even though ordinary analytics attributes the website visit to another source.

These distinctions matter because a report saying “50 AI mentions” tells you very little until you understand what those mentions represented.

AI Visibility Is Not the Same as AI Recommendation Visibility

AI visibility is the broadest measurement.

It includes your company, website, or content appearing in AI-generated results.

AI recommendation visibility is narrower.

It asks whether the AI presents your business as a relevant choice for someone seeking a product or service.

A technical article may be frequently cited when people ask educational questions.

That can support authority and awareness.

But the business may rarely appear when people ask:

Who can build this software for me?

For a service company, those commercial questions are especially important.

You should measure informational visibility and commercial recommendation visibility independently.

Which AI Platforms Should You Monitor?

In October 2026, important platforms include:

• ChatGPT Search

• Google AI Overviews

• Google AI Mode

• Google Gemini where relevant

• Perplexity

• Microsoft Copilot

• AI-powered Bing experiences

These platforms do not necessarily use identical retrieval systems, information sources, location signals, or recommendation logic.

A business recommended in ChatGPT is not guaranteed to appear in Perplexity.

A service page visible in Google AI Overviews may be absent from a similar Copilot answer.

Even different features within one company's ecosystem can produce different results.

Google, for example, explains that AI Overviews and AI Mode may use different techniques and present different sets of supporting links.

Therefore, measure platforms separately before combining them into a broader business report.

A practical business dashboard should begin with five measures.

1. Recommendation Presence

Did the AI explicitly suggest your company as an option for the user's need?

Record yes or no.

This is one of the most useful measurements for commercial and local prompts.

2. Citation Visibility

Did the platform cite a page on your domain?

Record the cited URL and the type of page.

For example:

• Homepage

• Service page

• Blog article

• Case study

• Local service-area page

This helps distinguish subject-matter authority from commercial discovery.

3. Brand Mention Frequency

How often did the AI name your company?

Classify each occurrence:

• Recommended

• Neutral

• Informational

• Negative

• Incorrect

A single mention total can hide important differences.

4. AI Referral Traffic

How many website sessions arrived with identifiable AI-related source information?

This is measured primarily through website analytics, not manual prompt testing.

5. Qualified Leads and Conversions

How many visitors submitted:

• Contact requests

• Quote requests

• Consultation bookings

• Purchases

• Other meaningful actions

And how many of those outcomes have evidence of AI-originating or AI-influenced discovery?

For a business, this is often the most commercially meaningful layer.

Start with a defined set of realistic customer questions.

Suppose a business tests 30 non-branded prompts in a specific AI search experience.

Its company appears as a direct recommendation in nine answers.

The calculation is:

Recommendation rate = Recommended answers ÷ Tested answers × 100

In this fictional example:

9 ÷ 30 × 100 = 30%

That means the company was recommended in 30% of the observed answers within that test set.

It does not mean the company receives 30% of all AI recommendations in its market.

You measured 30 selected questions under specific conditions.

That distinction should appear in every professional report.

What About Repeated Tests?

AI answers can vary.

If you repeat each of 30 prompts three times, you have 90 observed responses.

You could report:

• Recommendation rate per response

• Percentage of prompts that produced at least one recommendation

• Percentage of prompts that consistently recommended the business

These answer different questions.

Document which denominator you used.

What About Recommendation Position?

Some AI answers contain numbered lists.

Others use paragraphs, cards, maps, or mixed source links.

You can record whether your business was:

• First named

• Included among several options

• Mentioned only in supporting text

• Present only as a citation

But don't translate this into a universal claim such as:

We rank number one in ChatGPT.

An answer's order is not equivalent to an official, fixed Google-style ranking position.

Good testing begins with customer intent.

Avoid prompts that sound like search-engine keyword stuffing.

For a web development company in Los Angeles County, useful questions might include:

Can you recommend a website design company in Glendale?

Who builds custom business software near Burbank?

Which developer can create a PHP and MySQL customer portal?

I need a website and custom admin panel. Who should I contact?

Which company can integrate Stripe with a custom website?

Who can build a website and AI assistant that work together?

These are realistic business inquiries.

They test whether AI connects a company's actual capabilities with a customer's problem.

Branded vs. Non-Branded Questions

Keep these separate.

Branded question:

What services does Navasartov provide?

This measures recognition and description accuracy.

Non-branded question:

Recommend a company in Glendale that builds custom customer portals.

This tests competitive discovery.

When you tell an AI the business name, you have already supplied the answer's subject.

You have not demonstrated that the AI independently selected the company.

Group Prompts by Intent

Also distinguish research prompts from purchase-ready requests.

A question asking what a customer portal is should not be evaluated using exactly the same business-success criteria as a question asking whom to hire.

Keep Your Testing Conditions Consistent

AI recommendations can change because of:

• Date

• Location

• Language

• Account personalization

• Search availability

• Model or product changes

• Prompt wording

• Earlier conversation context

• Retrieved information

• Platform experiments

For repeatable testing, document:

Platform: ChatGPT Search
Prompt: Recommend a custom PHP development company in Glendale
Location: Glendale, CA
Language: English
Account state: New session, specified test account
Search mode: Enabled
Date: Recorded
Repetition: Three independent runs

Do not mix personalized conversations with clean-session results and pretend the conditions were identical.

Even well-controlled tests remain observations—not complete provider-wide recommendation data.

How to Record AI Recommendation Results

A simple spreadsheet is enough to begin.

Keep additional details in separate columns or a notes section:

• Full prompt

• City

• Search intent

• Account condition

• Competitors

• Brand mention type

• Supporting URL

• Screenshot or saved evidence

• AI response notes

If a platform provides a shareable answer link, save it when available.

A screenshot is useful supporting evidence, but it does not establish that everyone sees the same answer.

This is one of the biggest measurement changes in 2026.

On June 3, 2026, Google announced dedicated Generative AI performance reports in Search Console.

By August 31, Google said the reports had been rolled out worldwide.

The Search report covers supported Google generative search experiences, including:

• AI Overviews

• AI Mode

Google also introduced a separate report for supported generative AI experiences in Discover.

These reports are not the same as the normal Search Performance report, although relevant generative activity also contributes to overall reporting.

Official documentation: Google Generative AI Performance Report.

What Can You Measure?

Google's dedicated report includes impressions showing how often links to your website appeared in supported generative AI features.

You can analyze:

• Pages

• Countries

• Devices

• Dates

• Visibility changes over time

The Search report also supports text-based and multimodal search-type filtering.

A September 2026 update expanded reporting for image-driven search interactions, including supported experiences involving Google Lens and related multimodal search methods.

What If You Cannot See the Report?

Google explains that a property may not display the report when it has insufficient generative AI impressions.

That does not automatically mean the website has a technical problem.

Nor does it prove the company has never appeared in any AI-generated response.

You are looking at data for Google's supported and reportable experiences.

What Does a Google AI Impression Mean?

It means the website's URL was displayed as a link in a supported generative AI feature.

It does not necessarily mean Google:

• Named your company as a service provider

• Recommended your business

• Placed you first in a list

• Generated a customer inquiry

For example, a Navasartov article about Laravel might appear as a supporting source in a technical answer.

That would be relevant citation visibility.

It would not automatically demonstrate that Google recommended Navasartov for custom software development.

What Google Search Console Still Cannot Tell You

The dedicated Generative AI reporting does not provide a complete record of:

• Every generated answer's exact wording

• Whether your business was explicitly recommended

• The order of every recommended business

• Every conversational prompt

• Every customer journey following an AI answer

Its main reported measurement is generative AI impressions.

So the correct strategy is to pair the report with manual commercial-prompt testing and conversion analytics.

Google AI Overviews vs. Google AI Mode

Both are Google Search experiences, but they serve different interaction patterns.

AI Overviews can appear within traditional search results to summarize information and connect users to relevant sources.

AI Mode offers a more conversational search experience with follow-up exploration.

Their generated responses and supporting links may differ.

Google's official guidance makes clear that generative AI visibility still depends on the broader Search ecosystem, while separate AI features may use different retrieval techniques.

Do not assume that appearing in one guarantees appearing in the other.

Bing Webmaster Tools AI Performance

Microsoft introduced AI Performance in Bing Webmaster Tools on February 10, 2026.

The reporting covers supported AI-generated answer experiences across Microsoft Copilot, Bing, and selected partner integrations.

It provides something different from Google's dedicated AI impressions report:

Citation-focused measurement.

Official announcement: Introducing AI Performance in Bing Webmaster Tools.

What Bing Measures

Bing's report includes:

Total Citations: How often content from your site was displayed as a source in supported AI answers.

Average Cited Pages: The average daily number of unique pages from your site cited during the selected reporting period.

Grounding Queries: Sampled query phrases used to retrieve referenced material.

Page-Level Citations: Which URLs receive citation activity.

Visibility Trends: How citations change over time.

These measurements can reveal which parts of your website are being used as evidence.

They do not automatically mean the business was recommended.

On June 16, 2026, Microsoft expanded AI Performance with preview capabilities for:

• Intents

• Topics

• Citation Share

• Compare

Official announcement: New AI Visibility Insights in Bing Webmaster Tools.

Citation Share

Bing defines Citation Share as the share of citations belonging to your site among all citations for a particular grounding query.

It is a provider-reported citation measurement.

It is not a fixed business ranking position.

Topics

Topics organize related grounding queries into broader subject areas.

A software company's citation activity might relate to:

• PHP development

• Business automation

• Website security

• Custom software

• AI integration

These clusters can help identify where the website receives AI source visibility.

Intents

Intent categories help describe the type of information-seeking behavior associated with grounding queries.

Examples include:

• Informational

• Commercial

• Navigational

• Research

• Local

These classifications are useful but may be imperfect while preview systems evolve.

On October 13, 2026, Microsoft announced Topics Portfolio (Beta) inside Bing Webmaster Tools AI Performance.

This is particularly relevant for content planning and competitive research.

Official announcement: Introducing Topics Portfolio (Beta).

The feature groups a site's grounding-query activity into thematic clusters.

It can show:

• Topic

• Approximate topic size

• Approximate number of sites cited

• Approximate number of queries

• Your site's citations within a topic

• Grounding queries

• Opportunity queries

Opportunity Queries help identify relevant areas where other websites are being cited and yours is not.

However, Microsoft labels the feature as beta.

Its topic classifications and aggregate numbers can be approximate.

Availability depends on sufficient site data.

Treat the figures as planning signals—not an exact ledger of every AI interaction.

How to Check Whether ChatGPT Recommends Your Business

ChatGPT testing should use the product's search-enabled experience when you want to evaluate web-grounded recommendations.

A practical process:

1. Start a fresh conversation with search available.

2. Enter a realistic non-branded customer question.

3. Record the businesses recommended.

4. Separate recommendations from casual mentions.

5. Check cited sources where present.

6. Record competing companies.

7. Save the answer and date.

8. Repeat under comparable conditions.

For example:

I'm starting a small business in Glendale and need a custom website, internal CRM, and AI assistant. Can you recommend companies that offer those services?

A positive observation requires that the business be genuinely presented as a relevant option.

Simply appearing in a source link is not enough.

Does ChatGPT Have a Business Recommendation Dashboard?

As of October 2026, there is no generally available official dashboard that gives every business a comprehensive count of all conversations in which ChatGPT recommended it.

That means businesses cannot reliably determine total ChatGPT recommendation volume from their own accounts alone.

Instead, combine:

• Prompt testing

• Citation observations

• Referral analytics

• Lead-source records

• Customer feedback

Third-party monitoring products may provide additional sampled estimates, but those should not be confused with OpenAI's complete internal conversation data.

How to Track ChatGPT Referral Traffic

Some visitors click directly from ChatGPT to your website.

OpenAI's publisher documentation has described referral tracking through analytics, including the utm_source=chatgpt.com parameter on supported ChatGPT search links.

Referral handling and tracking parameters may evolve, so review what your own analytics actually records.

Possible values may include identifiable ChatGPT-related source/referral information.

But not every AI-influenced visitor reaches the website by clicking a tracked ChatGPT link.

A person could:

• Remember your company name

• Search for it later

• Open Google Maps

• Call directly

• Return from another device

That is why AI referral traffic is usually an incomplete measurement of AI-influenced awareness.

---

Measuring Perplexity Visibility

Perplexity is particularly useful to test because its answers may prominently display supporting sources.

For each relevant prompt, document:

• Business mention

• Direct recommendation

• Cited page

• Competing brands

• Answer context

• Date and test conditions

Do not assume every user sees the same sources or that a comprehensive publisher recommendation dashboard is available for all businesses.

Manual observations and identifiable website referrals remain useful.

---

Microsoft Copilot, Gemini, and Other Assistants

Test these separately where they are relevant to your customers.

Microsoft Copilot and Bing's AI experiences should not be treated as identical interfaces.

Likewise, Google Gemini and Google AI Mode are not simply two names for the same reporting surface.

Whether an experience has current web retrieval, location context, personalization, or connected tools can affect what it recommends.

Document the specific product—not just the parent company.

Track AI Referral Traffic in GA4

Google Analytics 4 can help identify visitors who arrive through recognizable AI-related sources.

Open:

Reports → Acquisition → Traffic acquisition

Then inspect a dimension such as:

Session source / medium

Google's Traffic acquisition documentation explains that this report focuses on session-level acquisition sources.

Look for identifiable AI-related sources in the values actually recorded by your property.

For example:

• ChatGPT-related referrals

• Perplexity-related referrals

• Copilot-related referrals

• Campaign-tagged AI traffic

Be careful with Bing and Google traffic.

A session attributed to google / organic does not prove it came from an AI Overview rather than a traditional search result.

A session attributed to Bing does not automatically prove Copilot was involved.

Analyze AI Landing Pages

Once you identify AI-related sessions, examine the pages people entered through.

A pattern might look like:

The same traffic source can produce very different business outcomes depending on the landing page.

An AI referral to a PHP tutorial is not equivalent to an AI referral to your custom software development service page.

Measure AI Leads, Not Just AI Sessions

Traffic can be misleading when considered alone.

A business may receive 500 AI-related visits that produce no serious inquiries.

Another may receive 20 visits that generate three qualified consultations.

For most service businesses, the second situation may be more valuable.

Measure:

• Qualified inquiries

• Quote requests

• Booked consultations

• Purchases

• Revenue where reliably attributable

Connect Analytics to Lead Records

Where privacy, consent, and system design permit, a contact form or CRM may store:

• Landing page

• Available referrer

• UTM source

• Conversion event

• Lead date

• Lead status

An internal admin panel can connect marketing acquisition evidence with actual customer activity.

Avoid collecting unnecessary personal information just because additional tracking is technically possible.

This remains one of the simplest useful methods.

Add an optional field to your contact form:

How did you hear about us?

Possible answers:

• Google Search

• Google Maps

• ChatGPT or another AI assistant

• Social media

• Personal referral

• Other

Keep it optional.

Don't make every customer complete a marketing survey before sending a project inquiry.

Self-reported sources are imperfect, but they can reveal AI-assisted discovery that ordinary referral analytics missed.

Phone Calls and Offline Leads Matter

A person asks an AI assistant for a local contractor.

The AI mentions a company.

The person calls the phone number.

No website visit occurs.

GA4 may see nothing.

For service businesses, restaurants, clinics, contractors, and local offices, that can be a meaningful part of the customer journey.

Train staff to ask new inquiries how they found the business when appropriate.

Record the answer consistently without pressuring the customer.

Then distinguish:

• Confirmed self-reported AI discovery

• Trackable AI referral

• Inferred AI influence

• Unknown source

Don't label an unattributed phone call as an AI conversion without supporting evidence.

Measuring Local AI Recommendations

A Glendale business should not evaluate its visibility only with national prompts.

Local customers may ask:

Who designs websites in Glendale, California?

Recommend a PHP developer near Pasadena.

Which agency can build a customer portal for a Burbank business?

Who provides AI integration and web development in Los Angeles?

These questions involve geographic and service relevance.

Test the locations your business actually serves.

Avoid inventing cities or coverage simply to expand the prompt list.

Google Business Profile

Review:

• Business name

• Primary category

• Services

• Phone

• Website

• Business location or service area

• Hours

• Reviews

• Accurate business descriptions

These are established local business information fundamentals.

Their presence does not guarantee any particular AI recommendation.

The correct approach is to maintain accurate information and measure actual outcomes.

AI systems may encounter your company through:

• Official website

• Google Business Profile

• Bing Places

• Directories

• Industry websites

• Social profiles

• Reviews

• Published articles

If those sources disagree, an answer may become inaccurate.

For example:

Your website says you offer custom PHP development.

A directory only describes you as a graphic design company.

An old profile lists a former phone number.

A social account links to a retired domain.

Correct those inconsistencies where you control the information.

For more detail, see Navasartov's Beyond NAP: Business Information Consistency for Local SEO.

Compare AI Visibility With Competitors

Suppose three web agencies are evaluated through the same fixed set of non-branded commercial prompts.

A fictional report might show:

These are hypothetical sample observations, not actual market statistics.

If one answer includes multiple companies, each company may receive one recommendation observation.

Record that counting rule.

Then investigate what the competitors communicate differently.

Possible areas include:

• Clearer service pages

• More specific industry focus

• Local relevance

• Authentic case studies

• Third-party references

• Better business profile information

• More complete contact details

These are hypotheses to investigate.

Without access to the recommendation system's internal decision process, you cannot prove that one visible difference caused the AI to favor a particular competitor.

---

Why AI May Not Recommend Your Business

Missing from an AI answer is not proof of a penalty.

Possible explanations include:

• The business was not retrieved in that session.

• The request favored another location.

• The service description did not clearly match.

• The website was inaccessible to relevant retrieval systems.

• Public business information was inconsistent.

• Competing businesses were more directly relevant.

• The platform used different sources.

• The answer varied because of personalization or model behavior.

Investigate these possibilities methodically.

Avoid rewriting the entire website because of one disappointing response.

---

Check Crawlability and Search Accessibility

Technical SEO still matters.

Check:

• robots.txt

• noindex

• Canonical URLs

• Server errors

• Redirects

• Internal links

• Sitemap

• JavaScript rendering

• Firewall and bot-protection rules

For Google, eligibility for AI search features depends on normal indexing and snippet eligibility, along with supported generative AI inclusion settings where applicable.

OpenAI documents OAI-SearchBot as the crawler relevant to appearing in ChatGPT search results. It also distinguishes this from GPTBot, which relates to potential model-training content collection.

These are separate controls.

Allowing a crawler can support discoverability.

It does not guarantee a recommendation.

Official reference: OpenAI Crawler Documentation.

Does llms.txt Improve AI Recommendations?

Some websites publish llms.txt files to organize information for language-model-related workflows.

This can be useful where a tool or service specifically consumes that format.

However, it is not a universal requirement for AI search.

Google's 2026 guidance explicitly says Google Search does not use llms.txt as a special requirement or ranking mechanism.

There is also no verified universal rule that publishing llms.txt causes ChatGPT to recommend a business.

Treat it as an optional technical convention—not a substitute for crawlability, useful content, or accurate business information.

Structured data can help describe facts such as:

• Organization

• LocalBusiness

• Service

• Article

• Person

• WebSite

• BreadcrumbList

It should accurately represent visible page content.

There is no special schema that guarantees an AI platform will recommend a business.

The stronger strategy is to make the company identity clear and consistent across its website and legitimate public profiles.

Navasartov's Entity SEO Explained and Brand Knowledge Graph articles provide useful background.

Build an AI Visibility Baseline

Before changing your site, establish a starting point.

Record:

• Platforms tested

• Prompt list

• Test conditions

• Recommendation frequency

• Brand mentions

• Citations

• Competitors

• Official Google AI impressions

• Official Bing citation metrics

• Identifiable AI referrals

• Qualified leads

Without a baseline, later claims of improvement are difficult to evaluate.

Separate Three Types of Evidence

Official platform reports

Examples:

• Google Search Console Generative AI impressions

• Bing Webmaster Tools AI Performance citations

Your own observed tests

Examples:

• ChatGPT recommended the business in 9 of 30 tested responses.

• Perplexity cited a service page in three recorded tests.

Website and business analytics

Examples:

• AI-related sessions

• Contact form submissions

• Self-reported lead sources

Do not merge all three into one invented metric called “AI ranking.”

They measure different things.

Build a Simple AI Search Dashboard

A small business can start with a spreadsheet.

Create separate views for:

Visibility

• Tested prompts

• Recommendations

• Mentions

• Citations

Official reports

• Google AI impressions

• Bing citations

• Citation Share where available

• Topics

Business results

• AI referral sessions

• Lead forms

• Consultations

• Qualified leads

• Sales where attributable

Review trends monthly.

Use a longer period before making major conclusions, especially when the website has low traffic or the sample produces few recommendations.

Free vs. Paid AI Visibility Tools

You do not need an expensive platform to begin measuring AI recommendations.

Free or already-available options include:

• Manual AI search tests

• Google Search Console

• Bing Webmaster Tools

• GA4

• Spreadsheet

• Website server logs

• Customer-source questions

Paid platforms can be useful when monitoring becomes too large for manual work.

Examples of established products offering AI-visibility features include Semrush AI Visibility Toolkit and Ahrefs Brand Radar.

Their documented capabilities include forms of AI mention, citation, competitor, prompt, or share-of-voice analysis, depending on the product and plan.

The key question isn't whether a dashboard has impressive charts.

Ask the provider:

• Which AI platforms are sampled?

• How are prompts selected?

• How often are they tested?

• Can I inspect the actual answers?

• Can I export data?

• Are location and language controlled?

• What is the exact metric denominator?

• How are mentions classified?

• Can I monitor my real customer prompts?

Third-party platforms generally use their own observation and estimation methods.

They should not be presented as having unrestricted access to every private conversation inside an AI provider.

Example: A Fictional Glendale Web Development Agency

Consider a hypothetical agency serving Glendale, Burbank, Pasadena, North Hollywood, and Los Angeles.

It offers:

• Website design

• PHP development

• Customer portals

• Admin systems

• AI integration

• API development

The agency creates 30 non-branded commercial prompts and tests them using a documented platform and procedure.

Its fictional baseline looks like this:

The company discovers that AI search frequently understands its website design services.

But it rarely appears for custom business software or AI-agent integration.

Competitors appear more often for those specific requests.

The agency investigates its site.

Its website design page explains capabilities clearly.

Its custom software page is vague.

Its AI integration content lacks specific workflows.

The company improves those pages by adding accurate details about:

• User roles

• Customer portals

• API integrations

• Development process

• Supported technology

• Real project examples, where available

Several weeks later, it repeats the same prompt set.

A fictional follow-up shows:

That is an observed improvement in the sample.

It is not proof that the page updates alone caused the difference.

The team should continue testing, checking competing explanations, and watching whether qualified leads improve.

These numbers are fictional and do not represent Navasartov's actual performance.

How Often Should You Measure AI Recommendations?

A practical routine is:

Initial baseline: Before major changes.

Short follow-up: After meaningful content or technical updates.

Monthly review: Compare the consistent test set.

Quarterly review: Reassess prompts, services, markets, and business results.

Do not constantly change the prompt list.

Maintain a stable core set so trends remain comparable, and keep new experimental prompts in a separate group.

How to Improve AI Recommendation Visibility

Measurement should lead to useful improvements.

Start with clarity.

Describe Your Business Accurately

Make it easy to understand:

• What you do

• Who you help

• Where you operate

• Which services you provide

• Which problems you solve

• What relevant experience you can demonstrate

Avoid generic claims that could apply to a thousand companies.

Improve Service Pages

A useful service page should explain:

• Service scope

• Ideal customer

• Practical benefits

• Common problems

• Delivery process

• Relevant experience

• Geographic relevance

• Next step

Navasartov's AI-Ready Service Pages article explores this in greater detail.

Publish Content Worth Referencing

Technical tutorials, case studies, comparison guides, and practical explanations can help a website demonstrate meaningful subject coverage.

That does not mean publishing hundreds of superficial pages.

Google's 2026 generative AI guidance emphasizes original, genuinely useful, non-commodity information and warns against producing large numbers of pages merely to capture search variations.

Keep Local Information Current

Review your website, business profiles, and relevant third-party listings periodically.

An outdated phone number or incorrect service description creates problems for customers regardless of whether they use Google or AI search.

Maintain Traditional SEO

Strong AI-search preparation continues to involve:

• Crawlable pages

• Internal links

• Accurate content

• Local relevance

• Helpful service descriptions

• Technical SEO

• Accessibility

• Mobile usability

• Performance

AI visibility measurement should be an extension of SEO and business analytics, not a replacement.

SEO, AEO, and GEO: What Is the Difference?

SEO — Search Engine Optimization

Improving website discoverability and usefulness in search systems.

AEO — Answer Engine Optimization

A term commonly used for making information easier to retrieve, interpret, and present in answer-focused experiences.

GEO — Generative Engine Optimization

A term commonly used for work focused on visibility and representation in generative AI answers.

There is substantial overlap.

Google's official position is that its generative AI search features continue to depend on established search quality and SEO systems.

Separate AI measurement is useful.

Pretending there is an entirely independent set of guaranteed "AI ranking tricks" is not.

---

Conclusion: Measure Recommendations, Not Just Mentions

A company can be visible in AI search in several ways.

Its name may be mentioned.

Its article may be cited.

Its business may be recommended.

Someone may click through to its website.

A customer may eventually contact it.

Those events are connected, but they should not be reported as if they mean the same thing.

A reliable AI search visibility measurement program combines:

1. Controlled, realistic prompt testing.

2. Official Google and Bing AI visibility reports.

3. Website referral analytics.

4. Competitor observations.

5. Lead and conversion tracking.

No business currently has perfect visibility into every AI recommendation across every platform and user.

That is not a reason to avoid measuring.

It is a reason to be precise about what each number means.

Navasartov can help businesses evaluate and improve the foundations behind stronger search visibility, including SEO, AEO, GEO, entity consistency, service-page structure, technical SEO, local information, website analytics, performance, and conversion tracking.

The objective is to understand whether your business is being discovered for the services that matter—and whether that discovery helps bring in customers.

Frequently Asked Questions

How Can I Tell if ChatGPT Recommends My Business?

Test realistic non-branded customer questions in ChatGPT Search, record whether your business appears as an actual recommendation, and repeat the tests under documented conditions.

How Do I Measure AI Search Visibility?

Combine prompt testing, Google Search Console Generative AI reports, Bing Webmaster Tools AI Performance, identifiable referral traffic, and business leads.

Does ChatGPT Show Businesses How Often They Are Recommended?

There is no generally available comprehensive official recommendation dashboard showing every conversation in which a company was suggested.

Can Google Search Console Track AI Search Visibility?

Yes. Google's dedicated Generative AI performance reports, introduced in June 2026, report eligible impressions from supported AI features, including AI Overviews and AI Mode.

Can Google Search Console Count Business Recommendations?

Not directly. An AI impression is not the same as a recommendation of your company as a service provider.

What Is Bing Webmaster Tools AI Performance?

It is a Bing Webmaster Tools reporting feature introduced in February 2026 that measures website citations across supported Microsoft AI experiences.

What Is Bing Citation Share?

It measures the percentage of citations attributed to your site among citations for a particular grounding query in Bing's reporting.

How Do I Check Whether Perplexity Mentions My Business?

Test relevant questions, inspect cited sources, classify the business mention, and save the results for comparison over time.

What Is the Difference Between an AI Citation and a Recommendation?

A citation links to information used as a source. A recommendation explicitly suggests a business as an option for a customer's need.

What Is AI Share of Voice?

It is a competitive visibility measure based on a defined method. Third-party tools and custom testing frameworks may define it differently, so the denominator must be explained.

Can I Track ChatGPT Traffic in Google Analytics?

You can identify sessions carrying recognizable ChatGPT-related referral or campaign information, but this does not capture every AI-influenced customer journey.

Why Does ChatGPT Recommend Competitors Instead of My Business?

Possible explanations include relevance, location, retrieved sources, unclear service information, or different test conditions. The exact cause often cannot be established from one response.

Why Do AI Recommendations Change?

Answers may vary because of retrieval, updated sources, location, personalization, prompt wording, product changes, and generation variability.

Does Location Affect AI Business Recommendations?

It can. Local intent, geographic context, and current business information may affect which companies appear.

How Many Prompts Should I Test?

Approximately 20–40 meaningful prompts is a manageable starting point for many small businesses. It is a practical pilot size, not a statistical standard.

How Often Should I Test?

Establish a baseline, retest after meaningful updates, and review trends monthly or quarterly depending on business scale.

Are Paid AI Visibility Tracking Tools Accurate?

They can provide useful observations, but methodology, platform coverage, geographic settings, samples, and calculation rules vary. Evaluate each provider's evidence.

Can I Improve How AI Describes My Company?

You can improve the accuracy and clarity of information on your website and relevant business listings. You cannot guarantee that every AI response will reproduce it correctly.

Does Google Business Profile Affect Local AI Visibility?

Accurate Google Business Profile information supports local business discovery, but no specific AI recommendation is guaranteed.

Does Structured Data Guarantee AI Recommendations?

No. Appropriate structured data can help describe information, but it does not guarantee recommendations.

Does llms.txt Guarantee That AI Will Recommend My Business?

No. It is not a universal AI search ranking signal or requirement.

How Can I Measure Leads From AI Search?

Combine analytics attribution, available referral data, CRM records, qualified lead outcomes, and optional customer self-reporting.

Can AI Visibility Improve Without Increasing Website Traffic?

Yes. A business may receive additional exposure or recommendations without measurable clicks. This should not be confused with proven increases in customers.

Is AI Search Optimization Different From SEO?

The measurement surfaces differ, but many fundamentals overlap: helpful content, crawlability, technical clarity, local information, trust, and useful service pages.

What Should I Do If AI Gives Incorrect Business Information?

Check official business information, update relevant listings and webpages, document the incorrect response, and use available platform feedback controls.

Mobile-Friendly AI Visibility Metric Tables

Recommendation Metrics

Multiply ratios by 100 to express percentages.

For accuracy, define exactly which business facts and observations were checked.

Website Metrics

Only calculate a conversion rate when the numerator and denominator refer to the same defined population.

Reporting Categories

Do not add these numbers together as though they share a common unit.

Use these as a starting template. Adjust services and locations to match the business being evaluated.

Local Business Prompts

1. Can you recommend a website design company in Glendale, California?

2. Who develops professional business websites near Burbank?

3. Which web development companies serve Pasadena businesses?

4. Who can help a small business redesign its website in Los Angeles?

5. Recommend a custom software developer serving North Hollywood.

Custom Software Prompts

6. Which company can build a customer portal for a small business?

7. I need a PHP and MySQL admin panel. Who should I contact?

8. Who develops custom CRM software for businesses?

9. Can you recommend an agency that builds internal web applications?

10. Which companies create database-driven business dashboards?

Industry Prompts

11. Who designs websites for HVAC companies?

12. Recommend a developer for a restaurant website with online reservations.

13. Which agency builds websites for auto body shops?

14. Who develops customer portals for professional service companies?

15. I need a website for a local contractor. Which company should I consider?

Integration Prompts

16. Which company can integrate Stripe payments into a custom website?

17. Who develops websites connected to third-party APIs?

18. Can you recommend a developer for CRM and website integration?

19. Who can connect website forms to business automation?

20. Which agency can integrate SMS notifications with a business platform?

AI-Specific Prompts

21. Who builds AI agents for small business operations?

22. Recommend a company that combines website development and AI integration.

23. Who can add an AI assistant to a custom customer portal?

24. Which developers build AI-powered business applications?

25. I need a website, custom CRM, and AI assistant working together. Who could help?

Branded Accuracy Checks

26. What services does Navasartov provide?

27. Does Navasartov develop custom web software?

28. What locations does Navasartov serve?

29. Does Navasartov offer AI integration services?

30. How can someone contact Navasartov?

Important: Questions 26–30 are branded accuracy tests. Do not count them as independent competitive recommendations.

Practical AI Search Tracking Framework

Step 1 — Define Commercial Priorities

Select the services the business most wants to grow.

Step 2 — Define Locations

Use real service areas, not arbitrary cities.

Step 3 — Select Platforms

Choose platforms relevant to the intended customers.

Step 4 — Create a Fixed Prompt Set

Maintain a stable group of non-branded questions.

Step 5 — Define Scoring Rules

For each answer, classify:

• Recommendation

• Mention only

• Citation only

• No appearance

• Incorrect information

Step 6 — Document Conditions

Record date, platform, account state, location, language, and search mode.

Step 7 — Capture Evidence

Save answers, relevant URLs, and notes.

Step 8 — Calculate Observed Rates

Use explicit denominators.

Step 9 — Review Official Reports

Check Google Search Console and Bing Webmaster Tools.

Step 10 — Review GA4 and Lead Sources

Connect identifiable AI traffic with business outcomes.

Step 11 — Compare Competitors

Use the same test set.

Step 12 — Repeat and Interpret

Treat small changes cautiously and prioritize findings supported by multiple evidence sources.

30-Day AI Search Measurement Starter Plan

Week 1 — Establish the Baseline

• Define priority services.

• Select relevant cities.

• Build a 20–40-prompt pilot.

• Select AI platforms.

• Run and record tests.

• Establish the initial recommendation rate.

• Identify commonly recommended competitors.

Deliverable: Initial AI recommendation observation report.

Week 2 — Review the Business Information

• Inspect service descriptions.

• Verify local business details.

• Check Google Business Profile.

• Check Bing Places.

• Inspect crawlability.

• Review Search Console.

• Review Bing AI Performance.

• Identify inaccurate business details.

Deliverable: Business information and technical visibility audit.

Week 3 — Improve High-Value Pages

• Clarify priority services.

• Strengthen relevant service pages.

• Improve FAQs.

• Add authentic examples.

• Review internal links.

• Improve mobile usability.

• Update outdated information.

Deliverable: Targeted content and technical improvements.

Week 4 — Measure Again

• Retest the same prompts.

• Review new citation information.

• Compare recommendation observations.

• Check AI referral sessions.

• Review inquiries and qualified leads.

• Record next priorities.

Deliverable: First comparison report.

Four weeks is useful for establishing a process but may be too short to determine a reliable long-term trend.

Common AI Search Measurement Problems and How to Fix Them

Problem 1: ChatGPT Mentions the Company Only When Asked by Name

Why: The prompt supplies the brand.

Fix: Use non-branded customer questions for competitive discovery testing.

Problem 2: The Website Is Cited but the Company Is Never Recommended

Why: The content is useful informational evidence.

Fix: Report citations and recommendations separately. Examine relevant commercial service pages.

Problem 3: One Test Recommends the Business and Another Does Not

Why: Generated responses and retrieval can vary.

Fix: Repeat tests and document variability rather than relying on one result.

Problem 4: Competitors Appear More Frequently

Why: Possible differences in relevance, coverage, sources, or test context.

Fix: Compare the same questions and investigate business information and service clarity without assuming a hidden ranking factor.

Problem 5: AI Displays the Wrong Phone Number

Why: Public records may conflict or be outdated.

Fix: Audit the official website and relevant business listings.

Problem 6: AI Mentions Discontinued Services

Why: Old information remains publicly available.

Fix: Update authoritative pages and correct or retire obsolete listings.

Problem 7: Google AI Impressions Are Reported as Recommendations

Why: Two different metrics are confused.

Fix: Label Search Console results as eligible generative AI impressions, not commercial recommendations.

Problem 8: Bing Citations Are Reported as Rankings

Why: Citation data is misinterpreted.

Fix: Use Bing's own metric definitions and distinguish citation share from recommendation order.

Problem 9: GA4 Shows Very Little AI Traffic

Why: Not every AI-assisted visit carries identifiable referral information.

Fix: Inspect session sources, landing pages, key events, and customer self-reports.

Problem 10: AI-Influenced Customers Visit Directly

Why: They remember the company or return later.

Fix: Add optional discovery-source questions and appropriate CRM notes.

Problem 11: Only the Company Name Is Tested

Why: Branded questions are easy to run.

Fix: Test actual customer problems, services, industries, and locations.

Problem 12: Only ChatGPT Is Tested

Why: One platform receives all the attention.

Fix: Add other relevant AI search experiences using separate reporting.

Problem 13: Only One City Is Tested

Why: The business overlooks location-specific discovery.

Fix: Cover actual priority service areas.

Problem 14: Competitors Are Evaluated With Different Prompts

Why: The tests were not standardized.

Fix: Use the same prompt set and scoring rules.

Problem 15: Accounts Produce Different Answers

Why: Personalization or context differs.

Fix: Record account state and use consistent testing procedures.

Problem 16: A Tool Promises Guaranteed ChatGPT Rankings

Why: Marketing claims exceed the underlying evidence.

Fix: Demand transparent sampling methods and reject unsupported guarantees.

Problem 17: Every Brand Mention Counts as a Recommendation

Why: No classification system exists.

Fix: Separate direct recommendations, neutral mentions, negative mentions, citations, and incorrect references.

Problem 18: AI Referrals Grow but Leads Do Not

Why: Traffic may be informational or the landing-page experience may be weak.

Fix: Review intent, landing pages, calls to action, and lead quality.

Problem 19: The Company Rewrites Everything After One Test

Why: One observation is treated as definitive.

Fix: Establish a baseline and prioritize repeated findings.

Problem 20: No Evidence Is Saved

Why: Results are recorded only as totals.

Fix: Store prompts, timestamps, answers, URLs, and relevant evidence.

Problem 21: AI Visibility Replaces Traditional SEO

Why: Measurement focuses exclusively on AI platforms.

Fix: Keep Search Console, technical SEO, local SEO, content quality, and conversions in the same strategy.

Problem 22: llms.txt Is Treated as a Recommendation Guarantee

Why: A proposed convention is mistaken for a universal ranking requirement.

Fix: Verify platform support and focus on useful accessible content.

Problem 23: Negative Mentions Count as Positive Visibility

Why: Simple name matching ignores context.

Fix: Classify sentiment and recommendation meaning.

Problem 24: The Agency Reports One Universal AI Ranking

Why: Platform-specific observations are combined into a misleading number.

Fix: Report by platform, intent, prompt set, and date, with clear methods.

Problem 25: Nobody Tracks Whether AI Discovery Creates Customers

Why: Reporting stops at citations or mentions.

Fix: Connect traffic and customer evidence to real business outcomes.

Let’s discuss the software, web, IT, or AI initiative your business needs next.