The short answer: measure three things - whether AI engines mention you, how you compare with competitors and how much traffic AI answers send you. You can check it free in about ten minutes with a fixed set of prompts, then read the rest in Google Analytics and Search Console. Build your test kit below.
In this guide:
- AI visibility is something you measure directly, and it's a different signal from your Google rankings.
- The free first-party layer (prompt tests, GA4, Search Console, Bing) gets most UK businesses a long way before any paid tool.
- The test-kit tool builds your prompt list, a scoring method and your GA4 tracking recipe in one go.
Written for UK marketers and business owners who want to know where they stand in AI answers - without buying a tool to find out.
More and more of your customers are asking an AI instead of scrolling a page of links. So the question quietly changed. It used to be "where do we rank on Google?" Now it's also "when someone asks ChatGPT or Google's AI for a firm like ours, do we come up - and what does it say about us?"
Most businesses have no idea. The uncomfortable finding across the 2026 studies is that plenty of brands already show up in AI answers, yet only a small minority measure it - so they're being described, recommended or skipped by AI every day and flying blind.
This guide fixes that. It's about measuring your AI visibility: the signals that matter, how to read them free, which tools are worth paying for and what to do when the answer is wrong. Actually earning more of those mentions is a separate job, and we've written it up in our answer engine optimisation guide - this one is about knowing where you stand.
What AI visibility is - and why it isn't your Google rankings
AI visibility is how often, and how prominently, an AI answer engine mentions or cites your brand when someone asks it a question you'd want to be the answer to. The engines that matter in the UK are ChatGPT, Google's AI Overviews and AI Mode, Perplexity, Gemini and Microsoft Copilot.
It's tempting to assume this is just SEO by another name, and it isn't. The two have visibly come apart: by early 2026 only around 38% of the pages cited in Google's AI Overviews still ranked in the top ten, down from roughly 76% a year earlier.
A page can sit on page two and be the source an AI quotes; a page can rank first and never get a mention.
That gap is the whole reason AI visibility needs measuring in its own right, rather than assumed from your rankings.
One more distinction worth drawing early, because it shapes everything below. This guide is about measuring where you stand - detecting whether you're mentioned, tracking it over time, comparing with rivals.
Improving it - earning more citations - is a related but separate discipline, and when a measurement points at a problem, the fix lives in our answer engine optimisation guide.
Keep the two jobs separate and both get easier. If you're still weighing AI search against classic search generally, our GEO versus SEO explainer covers that ground.
The four things to measure
"Are we in AI?" is too vague to act on. Break it into four measurable signals, because each answers a different question and needs a different method.
| What to measure | What it tells you | How to read it |
|---|---|---|
| Presence and inclusion rate | Whether you appear at all, and in what share of the questions that matter - the cleanest single number to start with | Manual prompt test, or a tool: the percentage of your tracked prompts where you're mentioned |
| Share of voice | How you compare with named competitors - your slice of all the brand mentions in your category | Your mentions divided by all brands' mentions across the same prompts, as a percentage |
| AI referral traffic | Whether that visibility actually sends you visitors, and what they do next | Google Analytics 4, with a custom channel group - and a realistic allowance for the traffic that hides in Direct |
| Sentiment and framing | Not just whether you're mentioned, but how - recommended, listed neutrally or described with a caveat | Read the actual answers; some paid tools score tone automatically, but a human read catches the nuance |
A word on the "mention" itself, because the vocabulary trips people up. A mention is your brand named in the answer text. A citation is the answer crediting your page as a source, usually with a link.
They don't always travel together - one 2026 study found only about one in five AI mentions carried a citation - so it's worth tracking both: mentions build awareness, citations send traffic.
You'll also see tools advertise a single "AI visibility score" out of 100. Treat it with care. Every vendor computes its own, with its own weighting, so no two are comparable, and a score means nothing without knowing how many prompts, which engines and how many runs sit behind it.
If you want one number you can reproduce, our test kit uses an open formula - and even then, read it as a trend on your own fixed set, never as an absolute grade.
Check it free in ten minutes: build your test kit
The fastest reliable way to see where you stand costs nothing: ask the AI engines the questions your customers ask, and write down what comes back.
The trick is testing the right questions in the right way. This tool builds a prompt set for your business, hands you a scoring method you can reproduce and gives you the exact recipe to track the traffic in Google Analytics.
Nothing is sent anywhere - it runs in your browser, and the prompts are yours to keep and re-run.
1. Your prompt set
Run each prompt in a logged-out or temporary chat, three times, and note whether you were not mentioned, mentioned, cited or linked. Score branded prompts separately - they measure recall, not discovery.
2. How to score it
For each prompt on each engine, give it 0 (not mentioned), 1 (mentioned), 2 (cited) or 3 (linked). Your inclusion rate is the share of cells where you scored at least 1. Your visibility score is your total points divided by the maximum possible, as a percentage. Enter your tallies below once you've run them.
3. Track the traffic in GA4
In Google Analytics 4, create a custom channel group, match the session source against this regex, and move the group above Referral. It catches the AI referrals GA4's default channel misses:
Remember: well over half of AI visits arrive with no referrer and land in Direct, so treat this as a floor. Don't add bare openai.com or bing.com - Copilot arrives as bing.com and would pollute your organic search.
How the score is worked out
Each prompt-and-engine pair scores 0 to 3 on the ladder above. The visibility score is your total points divided by the maximum (three points per cell), as a percentage; the inclusion rate is the share of cells where you appeared at all; share of voice is your mentions as a percentage of all brands' mentions. Branded prompts are scored separately, because being found for your own name measures recall, not discovery. It's an open, reproducible figure for tracking your own trend - not a certified benchmark, and not comparable with any tool's proprietary score.
Prefer to do this by hand, or have JavaScript switched off? Here's the method:
- Write 10-15 questions a customer would ask - mostly non-branded, e.g. "best [what you do] in [your area]", "how do I choose a [what you do] provider?", plus a few branded ones like "is [your brand] any good?"
- Run each in a logged-out ChatGPT, Perplexity and Google AI Overview, three times, scoring 0 (absent), 1 (mentioned), 2 (cited), 3 (linked).
- Inclusion rate = share of cells scoring 1+. Visibility score = total points ÷ (prompts × engines × 3) × 100.
- In GA4, add a custom channel group matching session source against:
chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.aiand move it above Referral.
Then re-run it on the same schedule and watch the trend.
The free first-party data you already own
Before you pay for anything, mine what you already have. Four free, first-party sources between them cover most of what a paid tool would tell you - and because it's your own data, it's the most reliable signal you'll get.
Google Analytics 4: your AI referral traffic
GA4 added a native "AI Assistant" channel in 2026, which is a start, but it has gaps: it misses Perplexity, and it doesn't capture the visits that arrive with no referrer.
The fix is the custom channel group the tool above generates - match the session source against the AI hostnames and place it above Referral so those visits get claimed before they fall into the generic Referral bucket.
The catch is large: well over half of identified AI visits arrive with no referrer at all - from mobile apps, copied links and privacy-stripped clicks - and land in Direct, where you can't attribute them cleanly.
So report three numbers rather than one: the AI traffic you can see, the Direct traffic that probably includes more of it and your Search Console AI impressions. Read together they triangulate; read alone, each misleads.
Google Search Console: the new generative-AI report
This is the one most businesses haven't noticed yet. In June 2026 Google added a generative-AI performance report to Search Console, showing how often your pages appear in its AI Overviews and AI Mode.
It launched in the UK first, under a Competition and Markets Authority requirement, and it's rolling out more widely - so if you have a Search Console account, it may already be there under Performance.
Know its limits before you lean on it. It's impressions-only - no clicks, no positions and none of the queries behind them - it blends AI Overviews and AI Mode into one figure, and it only covers Google, so it's silent on ChatGPT and Perplexity.
It also lives in the Search Console website rather than the API for now, so you can look but not easily export or automate.
Even so, it's the closest thing to an official, free scoreboard for your Google AI presence, and it's a genuinely useful trend line.
Bing Webmaster Tools: the Copilot blind spot
Microsoft added an equivalent AI performance report to Bing Webmaster Tools, and it does something Google's doesn't: it shows citation-level data for Copilot and Bing's AI answers.
That matters because Copilot traffic arrives in analytics looking like ordinary Bing search, so it's otherwise almost impossible to isolate. If any of your audience sits in the Microsoft ecosystem, this is the free report that covers the gap.
Here's the part that makes it, in one respect, more useful than Google's: Bing shows you the actual grounding queries - the real questions people asked that returned an answer citing you, each tagged with intent and topic. Google's report tells you which of your pages appeared; Bing tells you the language people genuinely typed. That's a direct read on real-world demand, and a strong steer on what to write next.
Branded search and your server logs
Two quieter signals round it out. A steady rise in branded searches in Search Console - people googling your name after an AI mentioned you - is one of the few ways to glimpse the "dark" traffic that never shows a referrer.
And your server or CDN logs show which AI crawlers visit: tools like Cloudflare's crawler analytics will flag GPTBot, ClaudeBot, PerplexityBot and the rest for free.
Two things to hold in mind there: a crawl is not a citation - most AI crawling is for training, not answering - and the bots that signal you're actually being used in answers are the retrieval ones (OAI-SearchBot, PerplexityBot and the user-triggered fetchers), not the training crawlers.
The tools: a straight, vendor-neutral comparison
Once measuring by hand becomes the bottleneck - too many prompts, too many engines, too often - a paid tool earns its place by doing it automatically and on a schedule.
What it's really selling is scale, not a signal you couldn't otherwise get. Here's the landscape, free options first. We don't resell any of these and get nothing if you buy them.
| Tool | Free option | What it measures | Paid from |
|---|---|---|---|
| Ahrefs (Brand Radar) | Yes - free one-off checker, no signup | Mentions across ChatGPT, Gemini, Perplexity, Copilot and AI Overviews | Paid tracking as an add-on (priced in £ for UK users; check current pricing) |
| Semrush | Yes - free AI visibility checker | Visibility score and competitor comparison | From ~$199/mo (~£147, in the wider toolkit) |
| Searchable (UK) | Yes - free report | Presence across ChatGPT, Google and Perplexity - the strongest UK-built free option | Paid tier (contact) |
| Rankscale / Otterly | Trial | Broad engine coverage, prompt tracking, citations | From ~$20-29/mo (~£15-21) - the cheapest ongoing trackers |
| Peec / SE Ranking | Trial | Visibility, citations, sentiment (Peec is EU-built, GDPR-friendly) | From ~€85/~$99/mo (~£73) |
| Profound | Demo | Page-level citations, sentiment, personas - enterprise depth | From ~$99/mo (~£73; more engines cost more) |
| Trakkr (UK) | Trial | Mentions, citations, perception - flat, published pricing in £ | From £79/mo |
Three points to weigh. Almost none of these price in pounds - Ahrefs and Trakkr are the exceptions - so most UK buyers are converting from dollars or euros and should check the live rate.
Every tool's headline "score" is proprietary and non-comparable, so pick on data source and engine coverage, not on whose number looks highest.
And there's a live debate about whether tools test real prompts people ask or synthetic ones they generate - the fair reading is that synthetic prompts are a decent proxy, not snake oil, but a tool that tracks the exact questions your customers ask is worth more than one that invents them.
Rather not wrangle a stack of tools? Our SEO retainers build in AI-visibility monitoring across ChatGPT, Perplexity and Google as standard - published pricing, no AI surcharge. See what's included and the prices.
What good looks like, and how often to check
The first question everyone asks is "what's a good score?" - and the real answer is that there's no audited industry benchmark yet.
The bands the tool vendors publish from their own data do broadly agree: a share of voice around 10-15% of the mentions in your category is a respectable base, 25-40% is strong and even category leaders rarely clear 60%. Business-to-business and considered-purchase categories tend to run higher than impulse consumer ones.
But chasing the absolute number is a trap, because it's meaningless without the prompts, engines and run-count behind it. The two readings that actually matter are your own trend on a fixed set of prompts and your position against the specific competitors you're tracking.
Both are relative, and both are real. A rising inclusion rate on the same twenty prompts is real progress; a shiny 47/100 from a tool you can't interrogate is not.
On cadence, weekly is the practical standard for most businesses, dropping to monthly once you have a baseline and only want the trend for reporting. Daily is worth it only during an active push or a competitive scare.
The reason not to over-measure is volatility: repeat the same prompt and only around 30% of brands persist run-to-run, so daily readings are mostly noise - and new content takes two to four weeks to start surfacing in answers anyway, so there's rarely much to see day to day.
Whatever cadence you choose, keep the prompts and the method fixed, or you're measuring your method, not your visibility.
When AI gets your brand wrong
Measuring often turns up something worse than absence: an AI confidently saying something untrue about you - an old price, a service you dropped, a merger that never happened, even "they've closed down". It's common enough to plan for.
One study of Google's AI Overviews found roughly 9% of answers contained a factual error, and AI engines cite sources incorrectly more often than you'd hope. Here's the practical response, and where measurement stops and improvement begins.
- Work out why. Run the false claim as a fixed prompt in a clean session and classify it: a stale source the AI is citing, confusion with a similarly-named business or a plain hallucination with no source. The fix depends on which.
- Fix your own source of truth. State the correct fact in plain, quotable text on your own site, with consistent Organization structured data and matching details across your main profiles. AI leans on what's clear and corroborated.
- Correct the high-authority sources it trusts. Wikidata is the quiet high-leverage one - it's treated as near-ground-truth and refreshes often - alongside Wikipedia (via the talk page, disclosing your connection) and any directories or review sites the AI cites. Never fake reviews; that's a different and worse problem.
- Use the in-product feedback as a backstop. ChatGPT's thumbs-down and "report", Google's AI feedback, Perplexity's report control. It's slow and unconfirmed, so treat it as a supplement while the source fixes do the real work. Under UK GDPR you also have a right to rectification of inaccurate personal data.
- Re-test on a schedule. Fixes propagate at different speeds: a day or two for live-retrieval answers, several weeks for knowledge-graph and Wikidata changes and up to months for anything baked into a model's training. Patience is part of the process.
That first step - detecting and classifying - is measurement, and it's this article's job. The steps that build lasting authority are optimisation, and they're covered properly in our answer engine optimisation guide. The durable lever, every time, is fixing the underlying sources - not arguing with the chatbot.
How Red Eagle Tech does it
We measure our own AI visibility the same way this guide describes, which is partly why we can write it.
That screenshot earlier is our real Search Console: over a recent three-month window, our pages drew around 114,000 impressions in Google's generative-AI features, and we can see exactly which ones the AI leans on - our SEO pricing guide alone accounts for close to half.
We watch the prompt tests, the GA4 channel and the Search Console trend the same way we'd set them up for a client.
If you'd rather not run all this by hand every week, it's part of what our SEO service does.
The £397 audit includes an assessment of where you currently stand in AI search, and the monthly retainers build AI-visibility monitoring in as standard - there's no AI surcharge, because measuring and improving AI presence is just part of modern SEO.
And because we build and fix websites too, when a measurement points at a technical cause, the same team can act on it.
Want to know where you stand in AI search? Our £397 audit covers your AI visibility alongside the rest of your SEO, and our retainers monitor it monthly - published pricing, no AI surcharge. See what's included and the prices, or talk to a UK SEO specialist.
Frequently asked questions
AI visibility is how often and how prominently an AI answer engine mentions or cites your brand when someone asks it a relevant question - across ChatGPT, Google AI Overviews, Perplexity, Gemini and Copilot. It's a different thing from your Google rankings: a page can sit outside the top ten and still be the source an AI quotes, and a page can rank first and never be mentioned. You measure it with three signals - whether you appear at all, how you compare with competitors and how much traffic AI answers actually send you.
Write down the 10 to 15 questions a customer would actually ask - mostly non-branded, like "best {your service} in {your town}" - then run each one in a logged-out or temporary chat, three times, and note whether you were not mentioned, mentioned, cited or linked. Do the same in Perplexity and Google AI Overviews. It takes about ten minutes and costs nothing. Our free test-kit tool below builds the exact prompt list for your business and gives you a scoring grid to keep.
Yes, largely. The free layer is genuinely useful: a manual prompt test you re-run on a fixed schedule, your AI referral traffic in Google Analytics 4 (with a custom channel group), Google's Search Console generative-AI report, Bing Webmaster Tools' AI performance report and your server logs for AI crawlers. Paid tools mainly buy you scale and automation - running hundreds of prompts across many engines on a schedule - rather than a signal you can't otherwise see. Most UK businesses should start free and only pay once measuring by hand becomes the bottleneck.
GA4 added a native "AI Assistant" channel in 2026, but it misses Perplexity and doesn't capture referrer-less visits. The reliable fix is a custom channel group: match the session source against the AI hostnames (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai and the rest) and move it above Referral. The catch is that well over half of AI-driven visits arrive with no referrer and land in Direct, so treat the visible number as a floor, not the full picture.
Because a lot of AI traffic strips the referrer. Clicks from mobile apps, from copied links and from tools that tag their links no-referrer arrive with nothing to identify the source, so GA4 files them under Direct. Independent estimates put the referrer-less share of identified AI visits at well over half. You can't fully recover it, but you can triangulate: watch the Direct channel for step-changes that line up with AI mentions, and read it alongside your prompt-test results and your Search Console AI impressions.
It's a report Google added to Search Console in June 2026 that shows how often your pages appear in its generative-AI features - AI Overviews and AI Mode. It launched in the UK first, under a Competition and Markets Authority requirement, and it's rolling out more widely. Two limits to know: it's impressions-only, with no clicks, positions or the queries behind them, and it blends AI Overviews and AI Mode together. It also covers only Google's own surfaces, so it tells you nothing about ChatGPT or Perplexity.
There's no audited industry benchmark yet, and any single 0-100 score is meaningless without knowing which prompts, engines and how many runs produced it. As a rough guide from tool vendors' own data, a share of voice around 10-15% of the mentions in your category is a solid base and 25-40% is strong, with even category leaders rarely above 60%. Far more useful than the absolute number are two things: your own trend on a fixed set of prompts and where you sit against the specific competitors you're tracking.
Weekly is the practical standard for most businesses, dropping to monthly once you have a baseline and just want the trend for reporting. Daily only earns its keep during an active push or a competitive scare. The reason not to obsess over daily readings is that answers are genuinely volatile - the same prompt can return a different answer most times you run it - so day-to-day wobble is mostly noise, and new content takes two to four weeks to start surfacing anyway.
They're a useful snapshot, with two caveats. First, every tool computes its own score with its own weighting, so numbers aren't comparable between tools - only a tool's own trend over time is meaningful. Second, most free checkers run a one-off sample of prompts you don't control, so they can miss the questions that matter to your customers. Use them for a quick read, but a fixed prompt set you run yourself, on a schedule, tells you more about the questions you actually care about.
The free checkers from Ahrefs, Semrush, Amplitude and UK-built Searchable cost nothing for a one-off read. Dedicated ongoing trackers start around $20-30 (roughly £15-22) a month at the entry end (Rankscale, Otterly), with mid-market tools like Peec, SE Ranking and Profound in the $85-100 (roughly £63-74) range and enterprise platforms well beyond. Almost none price in pounds - Ahrefs and Trakkr are the exceptions - so UK buyers convert from dollars or euros. Start with the free layer and only pay once you need scale.
First work out why: a stale source it's citing, confusion with a similarly-named business or a plain hallucination. Then fix the source of truth - state the correct fact in plain text on your own site with proper structured data, and correct the high-authority places AI leans on, especially Wikidata and any directories it cites. Use the in-product feedback (ChatGPT's thumbs-down, Google's AI feedback) as a slow backstop while the source fixes do the real work; under UK GDPR you also have a right to rectification. Then re-test on a schedule, because fixes take days for live sources and months for anything baked into training.
They overlap but they're not the same. Traditional SEO is about ranking in the list of blue links; AI visibility is about being the source an AI mentions or cites in its answer. The two have decoupled - by early 2026 only around 38% of the pages cited in Google's AI Overviews still ranked in the top ten, down from about 76% a year earlier. So you can't assume good rankings mean good AI visibility, which is exactly why it needs measuring on its own. Improving it is a related but separate job, covered in our answer engine optimisation guide.
Sources
- Google Search Central, Generative AI performance report for Search Console, published 3rd June 2026 - developers.google.com
- GOV.UK / Competition and Markets Authority, conduct requirements under the Digital Markets, Competition and Consumers Act 2024, published June 2026
- Google, GA4 "What's new" - the AI Assistant default channel, May 2026 - support.google.com
- Microsoft, Bing Webmaster Tools AI performance report, 2026 - bing.com/webmasters
- Ahrefs, AI Overviews citation study (Q1 2026 benchmark, ~38% of cited pages ranking top 10), published March 2026 - ahrefs.com
- AthenaHQ, State of AI Search 2026 - average brand mention rate and share-of-voice bands
- IAB UK / UKOM Ipsos iris, UK AI tool usage, 2026 - iabuk.com
- Semrush, AI search visibility research, 2026 - semrush.com
- AirOps and SparkToro, research on AI answer variability and run-to-run persistence, 2026
- Columbia Journalism Review, Tow Center study of AI citation accuracy, March 2025
- Newsweek / Oumi, analysis of factual errors in Google AI Overviews, April 2026
- Cloudflare, AI crawler and crawl-to-referral analytics, 2026 - cloudflare.com
