To track AI search visibility, test a consistent set of customer questions across your chosen AI search platforms, record brand mentions and source links, and compare results over time. Combine those observations with Google Search Console and website analytics to understand whether visibility brings relevant visitors and business enquiries.
A customer can discover your business inside an AI answer before visiting your website. Your company might appear as a recommendation, your article might become a cited source, or a competitor might receive the attention instead.
For a business owner, each situation raises a different question. Are people discovering us? Is the information accurate? Does that attention lead anywhere useful?
I recommend starting with a small measurement system that answers those questions. You can expand it when you know which information helps you make better decisions.
What is AI search visibility?
AI search visibility describes how your brand, products, services or content appear in AI-generated search answers. You can measure it through observed mentions, recommendations, citations and, where platforms provide them, impressions.
These signals have different meanings:
| Signal | What you record | What it tells you |
|---|---|---|
| Brand mention | Your company or product is named | The answer includes your brand |
| Recommendation | Your brand is presented as an option for the stated need | The answer gives your business a relevant role |
| Owned website citation | A source link points to your domain | Your website receives a visible reference |
| Third-party reference | Another website discusses your brand in a cited source | Someone else’s coverage may support discovery |
| Referral and outcome | A measurable visit leads to engagement, an enquiry or a sale | Some of the attention reaches your business |
A citation does not always name the business behind the page. A brand mention does not always include a clickable link. Record both so you can understand what actually happened.
If you are still developing your content approach, start with my guide to optimizing for AI search. This article focuses on measuring the results.
How to track AI search visibility in seven steps
1. Choose the business question and search platforms
Start with an outcome your team cares about. For example: “Do buyers looking for an ecommerce development agency see our company among the options?”
Choose two or three relevant platforms for your first round. You might include ChatGPT with web search, Perplexity and Google AI Mode. Add others when there is a business reason to track them.
Keep Google AI Overviews, AI Mode and the Gemini app separate in your records. They are different experiences. A result observed in one should not be reported as visibility across all three.
Also separate website accessibility from answer visibility. A crawler visiting your page is evidence of access; it does not prove the page appeared in an answer.
2. Build a fixed set of realistic customer questions
I suggest starting with 20 customer questions. That is a manageable planning choice for a small team, not a statistically representative sample of every possible search.
Use your sales conversations, support questions and existing search data to create the list. Include different stages of the buying decision:
| Question type | Example for a development agency | What you learn |
|---|---|---|
| Problem | Why is my online store slow on mobile? | Whether your educational content appears |
| Solution | Should a small retailer use Shopify or a custom store? | Whether your expertise supports a decision |
| Provider | Which agencies build ecommerce websites for small businesses in India? | Whether the brand enters a shortlist |
| Comparison | What should I compare when choosing an ecommerce development agency? | Which criteria and sources shape the answer |
| Branded | What services does [company name] provide? | Whether the platform describes you accurately |
Report branded and unbranded questions separately. When you include your company name in a question, a mention becomes much easier to obtain. That is useful for checking accuracy, but it is weak evidence of discovery by someone unfamiliar with you.
Keep a fixed core list for comparisons. Put new questions in a separate exploration list so changing the test does not look like improved performance.
3. Make each check repeatable
To track visibility in AI search engines consistently, record the conditions under which you obtained each answer.
Use the same question wording, language, target country and selected search mode. Start a fresh conversation and avoid earlier messages that tell the assistant about your company. Record the platform and model when the interface identifies them.
A clean conversation reduces some context effects; it does not remove every source of personalization or variation.
Run the core questions on a consistent schedule. Weekly checks are a practical starting point for a small site. For important questions, repeat checks on different days to see whether the result persists.
Repeated observations help reveal variation. They should not be treated as fully independent samples or proof that every customer sees the same answer.
For Google AI Overviews, record whether an overview appeared. “No overview displayed” is a valid search observation. A failed page load is a collection error. Those should never share the same status.
4. Save the answer and the evidence behind it
Create one spreadsheet row per question, platform and run. Keep these fields:
- Date, time and question ID.
- Exact question and intent category.
- Platform, search mode, country and language.
- Collection status and whether an AI answer appeared.
- Brand mentioned: yes or no.
- Brand recommended: yes, no or unclear.
- Your cited URLs and relevant third-party source URLs.
- Competitors named, with each brand counted once per answer.
- Accuracy notes and a saved answer or screenshot.
Write simple rules before collecting results. For example, count your accepted company-name variants, exclude unrelated businesses with the same name, and distinguish a passing mention from an explicit recommendation.
Check that a cited page actually discusses your brand before calling it a third-party brand reference. A competitor appearing beside an unrelated citation is not evidence that the citation supports that competitor.
Save the underlying responses. They let you revisit ambiguous cases when a dashboard number needs explaining.
5. Calculate metrics with clear denominators
A visibility percentage is only meaningful when you know what was counted.
Brand mention rate = valid answers mentioning your brand ÷ all valid answers in that test group × 100.
Owned citation rate = valid answers linking to your domain ÷ all valid answers in that test group × 100.
Recommendation rate = valid answers explicitly recommending your brand ÷ all valid answers in that test group × 100.
Calculate these separately by platform and question type. A branded accuracy check and an unbranded provider search should not have equal influence on your discovery headline.
For a hypothetical example, imagine 20 unbranded questions checked three times on one platform. You collect 60 valid answers. Your brand appears in 15 answers, and your website is cited in nine.
Your observed mention rate is 25%. Your owned citation rate is 15%. These figures describe that test set. They do not mean you reach 25% of the platform’s users.
If some checks fail, report attempted checks, valid answers and failures alongside the percentages. For successful Google searches without an AI Overview, retain a separate “no overview” count.
For that surface, calculate overview appearance rate as searches showing an overview divided by successful search checks. Then report citation rate among the overviews, with its denominator labelled. This separates fewer opportunities from weaker performance within available opportunities.
6. Compare competitors without confusing share of voice
Choose a fixed list of three to five relevant competitors. Count each brand at most once per answer, even if its name appears repeatedly.
For this method, observed share of voice equals your brand-answer appearances divided by the total brand-answer appearances for every brand in that fixed comparison group.
Suppose your brand appears in 15 answers and all tracked brands together account for 75 appearances. Your observed share of voice is 20%.
That differs from your mention rate because several brands can appear in one answer. Changing the competitor list also changes the denominator.
If a vendor reports share of voice or average position, read its definition before comparing it with your spreadsheet. There is no single universal “AI ranking” that all tools measure in the same way.
7. Connect visibility with website behaviour
Use website analytics to investigate the visits you can identify from AI platforms. In GA4’s Traffic acquisition report, use session-level traffic dimensions, such as Session source / medium, to examine where visits originate.
A practical workflow is to:
- Inspect your actual source values for recognizable AI services.
- Create a saved comparison or exploration using those observed values.
- Review landing pages, engagement and relevant key events.
- Track qualified enquiries or purchases where your implementation supports them.
Values containing domains such as chatgpt.com or perplexity.ai may help identify referrals, but inspect your own data before setting filters. Different links, apps and tracking conditions can produce different source values. OpenAI’s own publisher documentation states that ChatGPT search referrals automatically carry the parameter utm_source=chatgpt.com, which is a more precise value to filter on than the bare domain if your analytics setup preserves UTM parameters.
Some journeys lose referral information or end without a visit. Do not relabel unexplained direct traffic as AI traffic. Likewise, a google / organic session alone does not tell you whether someone used an AI Overview, AI Mode or a conventional result.
I would rather report a smaller, identifiable traffic segment than attach an AI label to visits we cannot explain.
Tools to track AI search visibility performance
Choose tools according to the evidence you need. A prompt tracker, a platform report and an analytics tool answer different questions.
| Tool or method | Useful for | Main limitation |
|---|---|---|
| Manual checks and a spreadsheet | A controlled starting sample and detailed accuracy review | Requires time and disciplined collection |
| Google Search Console | Google’s own reported generative AI visibility | Does not cover other companies’ AI platforms |
| Semrush Prompt Tracking | Monitoring selected questions and reviewing sources | Results depend on supported platforms and configuration |
| Ahrefs Brand Radar custom prompts | Tracking your chosen questions across supported AI platforms | A selected prompt set does not represent all user activity |
| Google Analytics 4 | Identifiable visits and on-site outcomes | Cannot observe every exposure or recover every missing referral |
Google Search Console: check Google’s own visibility data
Google’s Search Generative AI performance report announcement documents dedicated views of impressions in generative AI features, including AI Overviews and AI Mode. Its August 31, 2026 update states that the insights have rolled out worldwide.
The documented information includes impressions, pages, countries, dates and devices for Search. This visibility also remains included in overall performance reporting.
Use the available reports to identify pages receiving exposure and changes over time. Do not assume the dedicated view provides every click, ranking or conversion metric you would like to measure.
For the broader content strategy, see my guide to Google AI Overviews SEO.
Semrush: monitor a defined prompt set
Semrush Prompt Tracking supports selected prompts across ChatGPT, Google AI Mode and Gemini. Its documentation describes daily tracking, visibility and mention metrics, response snapshots and source reports.
This can suit a team that needs a recurring report without repeating every check manually. Before subscribing, confirm the supported countries, languages, prompt allowance and access to underlying answers for your plan.
Ahrefs: combine discovery with custom monitoring
Ahrefs’ custom prompt documentation explains how to select questions, supported platforms, locations and refresh frequency.
Custom monitoring helps you follow questions that matter to your business. Keep those results separate from visibility estimates based on a vendor’s broader question database. Both can be useful, but they measure different samples.
One asymmetry worth planning around: Google now publishes its own generative AI performance report directly to site owners, but OpenAI does not currently offer an equivalent public dashboard for ChatGPT. Its publisher access remains a private, partner-level arrangement rather than something any site owner can open. For ChatGPT specifically, a manual check or a third-party prompt tracker is currently the only way to observe what a customer sees, not a second-best option.
These tool descriptions are based on published documentation. They are not a controlled hands-on comparison or a promise that one platform captures every AI answer.
How to track brand visibility in AI search beyond a mention count
Read what the answer says about your company.
A recommendation for a service you no longer offer can generate poor enquiries. An incorrect location can confuse buyers. A pricing statement based on an old page can create expectations your team cannot meet.
For each brand-containing answer, check the business identity, services, location, important claims and linked sources. Label uncertain claims for review instead of automatically treating them as false.
I recommend giving factual accuracy its own dashboard row. A growing mention count should not hide a growing misinformation problem.
Also examine which questions produce recommendations. Visibility around a relevant buying decision may deserve more attention than a larger number of unrelated mentions. Keep any business-priority weighting visible and consistent so readers understand the score.
Turn your tracking report into a useful action plan
The report should tell you which page or business detail needs attention next.
| What you observe | What to investigate | Practical next action |
|---|---|---|
| Competitors appear; your brand rarely does | Missing topic coverage, unclear services or weak supporting evidence | Improve the relevant service page and answer genuine buyer questions |
| Your content is cited but the brand is absent | Whether the page clearly identifies its author and business | Clarify ownership and relevant expertise |
| Brand descriptions are inaccurate | Outdated pages or inconsistent third-party profiles | Correct information you control and request legitimate corrections elsewhere |
| Citations increase but visits remain low | Whether the answer satisfies the query without a click | Review question intent and the value offered on the destination page |
| Visits arrive but enquiries remain weak | Landing-page relevance, usability and the next step | Improve the page and verify enquiry tracking |
Google’s official guidance for generative AI search emphasizes useful content and established SEO foundations. It does not require a special writing formula or an llms.txt file for Google Search visibility.
My recommendation is to improve the substance behind the answer: clear service information, original examples, accurate comparisons and evidence readers can check. You can find related implementation guidance in my SEO guides.
Record every meaningful website change with its date. If visibility improves afterwards, report the association carefully. Platform changes, competitor activity and answer variation can also influence the result.
A simple 30-day tracking plan
During week one, select your platforms, write the core questions, define counting rules and check your analytics setup. Save the initial answers as your baseline.
During week two, repeat the same checks and investigate the most important gaps. Choose a small number of relevant pages to improve rather than rewriting the whole website at once.
During week three, publish those improvements and record what changed. Continue collecting the original questions under the same conditions.
During week four, compare valid sample sizes, mentions, citations, accuracy and identifiable business outcomes. Write three next actions, each with a responsible person and a review date.
Thirty days establishes a working process. It may not provide enough evidence to judge the long-term effect of your content changes.
Frequently asked questions
Start with manual checks, a spreadsheet, Google Search Console and GA4. Keep the question set small enough to repeat consistently. You still need time to collect responses and review their accuracy, and platform usage limits may apply.
Weekly checks are a reasonable starting point for a small business. Use additional checks for important changes or volatile questions. Review broader trends monthly, keeping the question set and collection conditions consistent.
There is no universal benchmark that applies across tools, platforms and industries. Compare your own baseline and relevant competitors using the same questions, definitions and conditions. Always show the underlying counts.
No. Referral traffic measures identifiable visits. It misses answers that mention your business without producing a visit, as well as journeys where referral information is unavailable. Combine it with answer observations and platform reporting.
No. Results depend on the audience, question intent, accuracy of the recommendation and the experience after someone reaches your business. Track qualified enquiries and revenue alongside visibility when you have reliable measurement.
Start with questions your customers actually ask
To track AI search visibility effectively, give every metric a clear purpose. Use mentions to assess inclusion, citations to understand source exposure, and analytics to investigate measurable business outcomes.
For a small business, I would start with 20 relevant questions, a few platforms and one consistent weekly review. Expand the system when the extra data changes a decision.
At Leelija Web Solutions, my company provides website, app and software development alongside SEO and social media marketing. That combination shapes my approach: search visibility should connect with a useful website and a clear next step for the customer.
Choose your first questions, save the evidence and let the findings guide the next improvement.