How-to guide · measurement
How to measure your AI search visibility without buying a platform
You cannot improve what you do not measure, and most businesses have no idea whether AI engines mention them. Here is a manual measurement system that takes two hours a month, the five numbers to track, and when it is worth paying for a tool.
Rankings told you where you stood. AI answers do not have positions, they have names, and either yours is there or it is not. Measuring that is a different job, and it has three parts: what the engines say (prompt tracking), who they send to you (GA4 and Search Console) and whether they can reach you at all (crawler logs). None of it needs a paid platform to start.
Step 1: Build your prompt set
Thirty to forty prompts across five groups, written the way customers speak, including your town or sector where relevant:
- Brand: "What is [company]?", "Is [company] any good?", "[company] reviews".
- Category: "best [service] in [town]", "[service] companies for small businesses in the UK".
- Problem: "my [thing] is broken, who do I call in [town]", "how do I [outcome]".
- Comparison: "[you] vs [competitor]", "[competitor] alternatives".
- Price: "how much does [service] cost in [area]".
Include five prompts you currently lose to competitors on purpose. Freeze the list for at least six months so the trend is comparable.
Step 2: Run them monthly across the engines
Same week each month, from a logged-out or fresh browser session. Engines: ChatGPT with search on, Perplexity, Gemini, Google AI Mode, Copilot. Optionally Claude. For each prompt and engine record in a spreadsheet:
| Column | Values |
|---|---|
| Date | |
| Engine | |
| Prompt | |
| Mentioned? | yes / no |
| Cited (your URL shown)? | yes / no |
| Position | 1st named, 2nd, 3rd, later, not named |
| Competitors named | list |
| Sentiment | positive / neutral / negative / wrong facts |
| Cited page (yours or theirs) | URL |
Two hours covers 35 prompts across five engines once you are practised. Screenshots into a dated folder help when someone questions the data.
Step 3: Calculate the five numbers
- Share of answer per engine: prompts where you are mentioned divided by total prompts.
- Citation rate: prompts where your URL is shown divided by total prompts.
- Competitor share: the same for your top three competitors.
- Accuracy rate: mentions where the facts stated about you are correct.
- Cited pages: which of your URLs appear, and which third-party pages appear instead of you.
Number 5 is the action list. Every third-party page cited in place of yours tells you what the engine could not find or trust on your site.
Step 4: Add the GA4 view
Set up the AI channel group from our GA4 guide, then each month record AI Search sessions, engaged sessions, key events, and the top ten landing pages by AI source. Landing pages by source is the nearest thing to a citation report you own, because it shows which pages engines are actually sending people to.
Step 5: Add Search Console
Google has been expanding Search Console reporting for its generative AI features through 2026. Record impressions and clicks attributed to AI features where available, and the queries and pages involved. Where a page ranks well but shows no AI-feature impressions, it is a candidate for the answer-first rewrite in our content guide.
Step 6: Check crawler access from your logs
Once a month, count requests and status codes for OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot, ClaudeBot, Bingbot and Googlebot on your key pages. A drop to zero or a rise in 403s means something changed at the edge, often a CDN update or a security plugin. Verify suspicious traffic against the published IP ranges rather than trusting user agent strings.
Step 7: Watch mentions off your site
Set up alerts for your brand name (Google Alerts, a Reddit search saved as a feed, LinkedIn notifications) and log new third-party pages that mention you. These are the sources engines draw on, and their growth usually precedes a rise in share of answer.
Step 8: Report it in one page
A monthly one-pager: share of answer per engine (with last month), citation rate, accuracy rate, AI sessions and conversions, top cited pages, top third-party pages cited instead of you, crawler access status, and the three actions for next month. That is the whole dashboard. It will get read.
When to move to a tool
The manual method breaks down above about 50 prompts, when you need daily sampling, or when you want automated competitor and sentiment tracking. At that point tools such as Profound, Peec AI, Otterly, Semrush's AI toolkit or Ahrefs Brand Radar pay for themselves in time. Keep the same prompt set so your history carries over.
Common mistakes
- Running prompts while logged in with personalisation on.
- Changing the prompt list every month, which destroys the trend.
- Measuring mentions but never asking which page was cited instead of yours.
- Reporting sessions only. Mentions without clicks still drive enquiries; the prompt data is how you see them.
Frequently asked questions
What is 'share of answer' or 'share of model'?
+
The percentage of a fixed set of relevant prompts for which an engine mentions or cites your business. If you run 30 prompts a month in ChatGPT and are named in 9, your share of answer in ChatGPT is 30 percent. Tracked per engine over time, it is the AI equivalent of rank tracking.
Do AI answers change between runs?
+
Yes. The same prompt can return different sources on different days and for different users. Run each prompt from a logged-out session, record the result, and read trends over months rather than single runs. Consistency across runs is itself a signal of strength.
When should I pay for an AI visibility tool?
+
When you track more than about 50 prompts, need daily rather than monthly data, want competitor share and sentiment automated, or need to report to a board. Tools such as Profound, Peec AI, Otterly, Semrush's AI toolkit and Ahrefs Brand Radar automate the prompt runs and the logging. Until then, the spreadsheet is adequate.
Sources
- DarwinApps: GA4 and CRM tracking for AI referrals · darwinapps.com
- Anagram: AI crawlers explained (verifying crawler visits) · anagram.ai
- Semrush: AI search trends (measuring presence alongside traffic) · semrush.com
Richard Daniel
Automation and Delivery Lead, Emerging Group
Richard leads automation and delivery across the Emerging group, working with EDP on client websites and with ETT on enterprise AI and process automation. He is the person who turns an audit finding into a working fix: crawler access, rendering, tracking, structured data and the plumbing that most marketing teams never see. He writes the technical guides on this site.
Read next
How to get indexed by Bing (because ChatGPT search leans on it)
OpenAI names Bing among the search providers behind ChatGPT search, and independent analyses in 2026 found most of ChatGPT's cited pages also rank in Bing's top results. Copilot is Bing. If your Bing coverage is thin, so is your ChatGPT visibility. Here is the fix.
How to run a 30-minute AI searchability audit of your own website
You do not need an agency to find out why AI engines are not recommending you. This is the exact half-hour checklist we run before any deeper work: access, rendering, entity, answers, schema, freshness and footprint. Most sites fail two or three of the seven, and they are usually fixable in a week.
How to check whether ChatGPT, Perplexity and Gemini can actually read your website
Before you spend a penny on AI search optimisation, find out whether the AI crawlers can even get in. This is a 20-minute check you can do yourself, and in our audits it fails more often than any content problem.