Why AI Visibility Matters
AI search is shaping recommendations and buying decisions. Visibility here means share of voice and future-proof demand.
Track AI Visibility
Monitor how often your brand appears in AI-generated answers from ChatGPT and Perplexity.
Stay Ahead of Competitors
See how your AI visibility compares to competitors and identify opportunities to improve.
Generate Optimized Content
Create content specifically designed to be picked up by AI engines and featured in answers.
Automate Schema Markup
Generate FAQ schema and structured data that helps AI understand and cite your content.
Choose Your Plan
Start tracking AI visibility today with flexible plans and no long-term commitment.
Ten AI tools, free with any account, no card and no time limit
Included for iDigitGroup clients
The free tools are free with no time limit and no card. The client tools are part of an iDigitGroup engagement rather than a separate subscription, so there is nothing to cancel and no trial to expire.
What the AI Visibility Checker measures
The iDigitGroup AI Visibility Checker is a free tool that asks AI assistants the questions your customers ask, then reports whether your brand was named in the answer. You give it a brand name and a few keywords. For each keyword it asks the connected assistants who the best providers are, reads the reply, and records whether you appeared, how far down, in what tone, and in what words.
This is a different question from the one an SEO tool answers. A ranking tool tells you where your page sits in a list of links. An AI assistant does not hand back a list. It writes a short answer and names two or three businesses inside it. You are either in that answer or you are invisible, and no ranking report will tell you which.
This page documents exactly how the check works: the question we send, how a mention is detected, how position and tone are read, and how the four parts of the score combine into the number on your dashboard. Nothing is withheld. You can reproduce the score by hand from this page.
Which assistants we check, and which we do not
We only report on platforms where an answer can actually be queried and verified. A number we cannot check is worth nothing to you, so we would rather show fewer platforms and stand behind all of them.
| Assistant | Status | What that means |
|---|---|---|
| ChatGPT | Checked | Answered by OpenAI GPT-4o where an OpenAI key is configured on the account, and by our built in model otherwise. Either way it is a live answer generated at the moment you run the check, not stored sample data. |
| Perplexity | Checked when connected | Answered by Perplexity's online Sonar models, which search the web before replying. This is the closest reading available to what a person sees in a live assistant. Where no Perplexity key is configured on the account, the Perplexity column is reported as not configured rather than counted as a miss. |
| Claude | Not tracked | Anthropic does not publish an interface that returns the answers real users are shown, so any Claude figure would be an estimate dressed up as data. We removed it rather than keep it on the page. |
| Google AI Overviews | Not tracked | AI Overviews are assembled per user, per query and per device, and Google does not expose them for querying. Nobody can measure this reliably today, ourselves included. |
Platforms are added as they become genuinely checkable. If you see a competitor's tool claiming daily Claude or AI Overview tracking, ask them which interface returns it.
How a single check runs, step by step
Every reading on your dashboard comes from the same five operations. They are simple on purpose, because a scoring method you cannot audit is a scoring method you cannot argue with.
1. The question we send
For each keyword you track, the tool sends the assistant a buying question in the form: what are the best {keyword} services or providers. It deliberately does not mention your brand, because naming you in the question would guarantee you appear in the answer and make the result meaningless.
2. The mention test
The reply is lower cased and searched for your brand name as plain text. If the string appears anywhere in the answer, the check is recorded as a mention. This is a strict test: a near miss, an abbreviation or a misspelling of your name does not count, which is deliberate, because an assistant that cannot spell your brand is not sending you customers.
3. The position reading
The tool finds the first line containing your brand. If that line begins with a number, as it does in a numbered list of recommendations, that number is your position. If it does not, the line's own index in the answer is used instead. Position one means you were the first business the assistant named.
4. The tone reading
The answer is scanned for a fixed list of positive words (best, excellent, top, recommended, leading, trusted, quality) and negative words (avoid, poor, bad, worst, not recommended, issues). Positive words with no negative words gives a positive reading, the reverse gives negative, and anything else gives neutral. This is a blunt instrument and we say so: the words are matched across the whole answer, not only the sentence about you.
5. The snippet
The tool stores the hundred characters either side of your first mention, so you can read the sentence in which the assistant described you rather than take our word for the tone.
How the visibility score out of 100 is calculated
Four components, each scored out of 100 and then weighted. The weights below are the ones the code applies, and the arithmetic is a weighted average rounded to the nearest whole number.
| Component | Weight | How it is calculated |
|---|---|---|
| Mention rate | 35 per cent | The share of all checks in which your brand appeared. Three keywords across two assistants is six checks, so four mentions is a mention rate of 66.7. |
| Average position | 30 per cent | Scored as 100 minus 15 points for every place below first, floored at zero. Position one scores 100, position three scores 70, position seven scores 10. Where you were never mentioned, average position is treated as ten. |
| Sentiment | 20 per cent | Positive mentions minus negative mentions, divided by total mentions, shifted onto a nought to 100 scale. All positive scores 100, an even split scores 50, all negative scores nought. With no mentions at all it sits at 50 rather than punishing you twice for the same absence. |
| Platform coverage | 15 per cent | The share of the assistants checked that named you at least once. Named on both of two assistants scores 100, named on one of two scores 50. |
What the tool tells you at each threshold
Mention rate below 50
You are absent from more than half the answers
The dashboard flags that mention rate is below average and points you at building more authoritative content. In practice this is a content depth problem, not a tracking problem.
Average position worse than three
You are named, but late
The dashboard flags that you are not in the top positions when you are mentioned. Assistants tend to lead with the provider they can describe most specifically, so a vague positioning shows up here first.
Sentiment score below 60
The language around you is flat or negative
The dashboard suggests addressing negative feedback and surfacing evidence of results. Read the stored snippets before acting, because the tone reading is word based and can be blunt.
Score of 70 or more
Strong AI visibility
The dashboard confirms your position is strong and the job becomes maintenance: keep the content that earned the mentions current, and keep checking as the models are retrained.
The score describes how assistants answered at the moment you ran the check. Models are retrained and reweighted without notice, so treat a single reading as a snapshot and the trend across several as the real signal.
A worked example, so you can check our arithmetic
A clinic tracks three keywords and both assistants are connected, which gives six checks in total. The clinic was named in four of them, at positions two, three, five and two. Three of those four mentions read as positive and none read as negative.
Mention rate: 66.7 out of 100
Four mentions out of six checks. Weighted at 35 per cent, this contributes 23.3 points.
Average position: 70 out of 100
Positions two, three, five and two average to three. 100 minus 15 times two gives 70. Weighted at 30 per cent, this contributes 21 points.
Sentiment: 87.5 out of 100
Three positive and no negative across four mentions. Weighted at 20 per cent, this contributes 17.5 points.
Platform coverage: 100 out of 100
The clinic was named at least once on both assistants checked. Weighted at 15 per cent, this contributes 15 points.
23.3 plus 21 plus 17.5 plus 15 gives 76.8, which the dashboard rounds to 77. The clinic is above the strong visibility threshold, and the cheapest improvement available is position rather than mention rate: moving from an average of three to an average of two would add another 4.5 points on its own.
This tool, the AEO checker and the SEO audit
We run three free tools and they answer three different questions. Running the wrong one is the most common reason people conclude a site is fine when it is not.
| Tool | The question it answers | Use it when |
|---|---|---|
| AI Visibility Checker | Am I being mentioned | You want to know whether assistants name your business when someone asks who the best provider is. Ongoing tracking across keywords and assistants. |
| AEO Checker | Is my page citable | You want to know whether one specific page is built so an assistant can read, trust and quote it. A one off diagnostic scored out of 100 against 18 checks. |
| SEO Audit | Is my site technically sound | You want the classic crawl: indexability, speed, structure, thin pages, internal links. The foundation the other two sit on. |
How to improve your AI visibility score
In order of how much difference each step makes, based on what moves the four components above.
- 1
Track the questions buyers actually ask
Start with three to five keywords that describe what you sell and where, phrased as a buyer would say them. Tracking a phrase nobody asks produces a low score that means nothing.
- 2
Publish the comparison page nobody wants to write
Assistants name providers they can describe specifically. A page that states who you are for, what you charge, what you do not do and how you differ gives a model something concrete to repeat. Vague positioning is the single most common cause of a low mention rate.
- 3
Make your entity unambiguous
One consistent business name, one address, one set of structured data across your site, your Google Business Profile and every directory. If the model cannot resolve which business you are, it will name one it can.
- 4
Get cited somewhere the models read
Mentions in the places assistants lean on carry more weight than another page on your own site. Industry press, professional directories and genuine community discussion all count. Buying links does not.
- 5
Answer the question in the first hundred words
Put a direct, quotable answer near the top of the page, then support it below. A model that has to reconstruct your answer from four paragraphs will usually quote someone who stated it plainly.
- 6
Fix the tone that shows up in your snippets
Read the stored snippets. If an assistant describes you in flat or hedged language, that is usually because the evidence it found was thin. Published results, dated case studies and named credentials change the words it reaches for.
- 7
Re-check after every substantial change, and monthly regardless
The check runs when you open the dashboard, so a reading is only ever as current as your last visit. Models are retrained without notice, and visibility can move without you touching anything.
What this tool cannot tell you
Stated plainly, because the limits matter as much as the readings when you are deciding what to do next.
- It reports what an assistant answered when we asked, not what it answered your customer. Assistants personalise on account history and location, so treat the reading as representative rather than exact.
- The mention test is an exact text match on your brand name. Abbreviations, misspellings and near misses are recorded as absences.
- The tone reading matches words across the whole answer, not only the sentence about you. An answer that calls a competitor the best option can read as positive. Always check the snippet before acting on sentiment.
- Position is read from a numbered list where one exists, and from the line order otherwise. In a flowing paragraph rather than a list, position is a rough ordering rather than a rank.
- The check runs when you open the dashboard. There is no overnight refresh, no scheduled crawl and no automated weekly email yet, so nothing is happening between your visits.
- Claude and Google AI Overviews are not covered, so a strong score here does not describe your visibility everywhere.
- A score is a measurement, not a diagnosis. It tells you that assistants are not naming you. Why they are not naming you is a content and authority question, and the improvement steps above are where that work starts.
Related pages
The rest of the answer engine optimisation toolkit and the reading behind it.
Free AEO Checker
Score one page out of 100 against the 18 checks that decide whether an assistant can quote it.
Free SEO Audit Tool
The classic technical and on page crawl, with every pass threshold published.
AI Search Optimisation
The service behind the tool, for when the work is bigger than a checklist.
SEO and AEO Glossary
Plain definitions of the terms used on this page.
AI Tool FAQs
Details about the AI Visibility tool and how to get value quickly.
Which AI platforms do you track?
ChatGPT and Perplexity. Where neither is connected the check still runs against our own built in model, so what you see is a real answer rather than sample data. We do not track Claude or Google AI Overview, because neither offers a way to query the answers real users get, and a figure we could not verify would be worth nothing to you. Platforms are added as they become checkable.
How often is the visibility data updated?
Every time you open the dashboard. The check runs there and then, so the result reflects what the AI answers at that moment rather than a stored overnight figure. There is no scheduled daily refresh and no automated weekly summary yet, so run it again whenever you want a fresh reading.
Is it really free, and what is not?
Ten tools are free with any account, with no card, no subscription and no expiry: mention tracking, the visibility score, the schema generator, the review responder, keyword clustering, the FAQ generator, the social post generator, the email template generator, the citation finder and the recommendations. Seven more, the content ideas, briefs, meta descriptions, title tags, blog outlines, Google Business Profile posts and content calendar, are part of an iDigitGroup client engagement rather than a separate plan.
How is the visibility score calculated?
Four weighted components. Mention rate is 35 per cent, average position 30 per cent, sentiment 20 per cent and platform coverage 15 per cent. Position scores 100 minus 15 points for each place below first. Sentiment is positive mentions minus negative mentions across all mentions, mapped onto a nought to 100 scale. The full method and a worked example are published on this page so you can reproduce your own score by hand.
What question do you actually ask the AI?
For each keyword you track we ask what the best services or providers are for that keyword. Your brand is deliberately left out of the question, because naming you would guarantee you appear in the answer and make the result meaningless. The reply is then searched for your brand name as plain text.
How is this different from your AEO checker?
The AEO checker answers whether one page is built so an AI can read, trust and quote it. This tool answers whether AI assistants are actually naming you when someone asks who the best provider is. A page can pass every AEO check and still not be mentioned, because being citable and being cited are different problems.
Why is my brand not being mentioned?
Usually because nothing published about you is specific enough to repeat. Assistants name the provider they can describe most concretely: who it is for, what it costs, what it does not do. A vague positioning, an entity the model cannot resolve to one business, and no citations on sites the models read are the three most common causes, in that order.
How often should I check?
Monthly is enough for most businesses, plus a check after any substantial change to your site or your presence elsewhere. A single reading is a snapshot, and models are retrained without notice, so the trend across several readings tells you far more than any one score.