AI visibility: how to measure whether AI search recommends you

A business can be named in an AI answer, linked as a source, or suggested as a place to buy. Google Maps results are a separate record. Keep those checks apart so you can see which kind of visibility changed.

How AI search visibility works

For a fixed set of customer searches, count how often an AI answer names the business or its website. A mention names the business. A citation links to a page. A recommendation puts the business forward as an option. Record each separately: a name without a link tells you something different from a cited service page or a direct recommendation.

A brand may appear in an answer without a link to its site. A page may be cited without the business being recommended. Save the wording, citation URL and recommendation for each prompt, then count each result against the same prompt set.

We treat this as a measurement question rather than one universal score. Our GEO vs SEO guide explains how search optimisation differs from work aimed at generated answers.

Where these answers appear

Check each search product separately. Google AI Overviews, Google AI Mode, ChatGPT search, Perplexity and Copilot can answer the same question differently. Keep the prompt, location and date with each result. Google includes AI Overview and AI Mode activity in Search Console's Web search data. Its Generative AI features report gives a separate view in the interface. See Google's AI features guidance and report announcement for details.

The Search Console Generative AI features report shows impressions for AI Overviews and AI Mode. It has no queries tab or API, and those impressions are also included in the usual Web search totals. Use SERP data and saved prompt runs for per-query checks. For other engines, keep a prompt log and review referral analytics. Microsoft Clarity documents AI bot and visibility features; check its AI Visibility overview before relying on a feature in your account.

Test 1: AI Overviews on 60 UK searches

AI Overviews appeared on 51 of the 60 searches. The query group made a clear difference. Forty-nine of 50 non-local searches showed an Overview. Two of ten service-plus-town searches did, while nine showed a local pack. We pulled the results through the DataForSEO SERP API on 25 September 2026.

Across the 51 Overviews, we counted 328 citations from 235 domains. Only 48% of cited URLs also ranked in the organic top 10 for the same search. The domain overlap was 56%. The number one organic result's domain appeared in 29 of 51 Overviews. A first-place result did not guarantee a citation. To check AI Overviews citations, compare each linked URL with the organic top 10.

YouTube was the most-cited domain, with 35 of 328 citations (10.7%). Government guidance and specialist sites appeared too. Check the source list for your own topic before assuming AI answers reuse only the highest-ranking pages.

Test 2: ChatGPT recommendations vs Google Maps

ChatGPT's Bristol recommendations seldom matched the Maps top three in our test. We sent 20 prompts, including "Which dentist in Bristol would you recommend? Give me the business names." The answers named 107 businesses. Six also appeared in the Maps top three for the same service.

ChatGPT named six businesses among the 60 Maps top-three places across those searches. For ten services, none of its recommendations appeared in the Maps top 20. We matched results by website domain or business name. The answer to the "SEO agency" prompt included Aristral once. That was one answer from one run, not a stable ranking.

We used gpt-4o-mini with web search through the DataForSEO AI Optimization API. This was not the consumer ChatGPT app. We ran each prompt once. Maps results came from one central Bristol coordinate, so businesses in outer districts may rank lower from that point. Treat the figures as directional. Our Google Maps ranking guide covers the local pack side.

How to check your presence in AI search

Start a spreadsheet before choosing an AI visibility checker. Keep prompt wording, date window and location the same each month. Save the answer or source links where possible. That makes changes easier to compare.

1. Write 20 to 30 fixed prompts from real customer questions. Include product, service, comparison and local intent where they fit your business.
2. Run the same wording in three AI search engines once a month. Record the engine, date, country, device and whether web search was enabled.
3. For each answer, mark whether the brand was mentioned, recommended or linked. Save every cited URL and note the order in which the answer presented options.
4. Check Google AI Overviews for your target searches and compare cited URLs with the organic results. Record when no Overview appears.
5. Use Search Console's Generative AI features report to review impressions from Google AI Overviews and AI Mode. The report is in the interface, has no queries tab or API, and its impressions are included in Web search totals. For query-level checks, compare SERP data with your prompt log. Google's AI features guidance explains its Web reporting.
6. Review server logs or analytics for visits and crawler activity. Clarity may help with AI bot visibility when its features are available. Check its current documentation and settings before treating those signals as complete.

A prompt log can include the question, engine, date, brand mention, recommendation, citation URL, answer position and notes. Count the prompts where you appeared and divide by the prompts you ran. Keep separate totals for mentions, citations and recommendations. Show the raw counts beside the percentage so readers can see what it represents.

When you report AI visibility tracking, say whether you mean manual checks, automated prompt runs, Search Console reporting or referral measurement. Each method answers a different question. Our llms.txt guide explains the file. Check crawler rules for each service you want to measure.

What should a useful score tell you?

A useful score has a clear method behind it. Ask which prompts, engines, locations and runs produced the number. Find out whether it counts brand mentions, source citations, recommendation position or a combined measure. Without those details, two scores may describe different results.

Use share of answers as a baseline: the number of relevant prompts where your business appeared divided by the prompts tested. Keep separate shares for citations and recommendations. Put the raw counts beside each percentage. Repeat the prompt set over time, and describe changes as observations rather than proof that one site edit caused them.

AI visibility tools: what to look for

Choose a tool whose measurement method you can inspect. Start with the output you need. A citation monitor may not track recommendations. A prompt tracker may not show whether a linked page ranks in organic search. Look for enough detail to check the results yourself.

Try a small prompt set before you settle on a tool. On the same day, compare its answer and citations with a manual run on the same engine. Ask how it handles missing answers, changed prompts and locations it cannot reproduce. If it only gives you a percentage, ask for the records behind it.

How to improve AI search visibility

Work on pages and sources that AI answers can use, then check whether they appear in your fixed prompt set. In our Google study, 48% of cited URLs also appeared in the organic top 10. The rest came from URLs outside those ten results. Organic rankings still gave pages a route into the answers.

1. Write pages that answer the questions customers ask. Put the answer near the top, then add details and evidence specific to your business.
2. Check citations for your topic. Improve pages that already earn relevant impressions. Fill gaps where an answer relies on a source your customers expect you to explain.
3. Make the business facts consistent across your site and the profiles customers use. Keep service names, locations and contact details accurate and easy to confirm.
4. Publish useful material on channels found in your source list. YouTube accounted for 35 of 328 citations in our Google sample. That figure applies to these queries on this date.
5. Review access rules for the search crawlers you want to reach. Our llms.txt explainer covers the file question; do not assume a text file overrides crawler access controls.
6. For local searches, improve your Google Business Profile and local citations alongside your website. In our Bristol test, AI recommendations and Maps results produced different shortlists. Check them separately.

For a wider plan, see AI Overviews optimisation. Choose actions tied to pages and sources your checks actually find. Then repeat the same prompts and note which answer, citation or recommendation changed.

How Aristral measures AI search

At Aristral, we track Google AI Overview citations from SERP reference data and run fixed prompt sets through language model APIs alongside Search Console. We record brand mentions separately from linked citations and local recommendations. Those records help us decide which pages or listings to review. Our Bristol test found different shortlists in AI answers and Maps results. Each needs its own check. The comparison was one run per prompt, so it does not establish a stable ranking. If you need help with this work, see our agentic SEO service and contact us.