How to choose an AI visibility tool — 7 criteria you should consider

Requirements specification over rankings: the seven questions you should ask an AI visibility tool before paying for it — engines, duplication, cited vs. recommended, performance, ownership, price and AI labelling.

Are you looking for a ranking of specific products? You’ll find it in our editorial top 5 list of GEO tools and on the comparison page. This article sets out the requirements: the seven questions you should ask about a tool before you pay for it — written by one of the industry players, with sources and disclaimers at the bottom.

An AI visibility tool measures whether your business is mentioned and quoted when customers search for ChatGPT, Claude, Gemini or Perplexity — and the best ones also fix whatever is blocking this on your own website. For a business without a marketing department, the choice is rarely determined by price. It is determined by whether the tool simply produces a report, or whether someone actually carries out the corrections afterwards.

Two figures explain why this category has emerged in the first place. Profound analysed 11.84 billion AI citations across the major AI engines and found that around 57 per cent of them pointed to the brand’s own website — so it’s your own site that matters most.

And the impact of your site has been measured, not guessed: the Princeton study behind the concept GEO (KDD ’24) found up to 40 per cent higher visibility in generative search engines when content was optimised with sources, statistics and citations. The question, therefore, is not whether there is anything to be gained. It’s who does the work — you or the tool.

What is an AI visibility tool?

It’s a tool that asks the AI engines the questions your customers ask, and records the results: are you mentioned by name, is your website used as a source, and who is mentioned instead of you? The category is new, the names are all over the place — AI visibility, GEO, AEO, LLM-SEO — and they all do the same job in practice.

The tools fall into three tiers. Most products on the market can handle the first two:

Tier	What it does	What you’re left with
Monitoring	Asks fixed questions to the AI engines and counts mentions and citations over time	A figure and a graph
Analysis	Scores your website based on what the machines read: readable HTML, structured data, llms.txt, robots.txt	A prioritised list of errors
Execution	Fixes the errors on the site — automatically, or as finished drafts for you to approve	Corrections that have gone live

The leap from the second to the third tier is what determines the value for a small business. A list of 40 technical errors is only beneficial if there’s someone to fix them. If you have neither a developer nor a marketing manager, the list will just sit there — and you’ll have paid to have a problem confirmed that you still have.

Why can’t Google Analytics measure this?

Because an AI response rarely results in a click. The customer receives the answer in the chat and moves on; only a fraction click through to the source on which the answer was based. So you’re being used without being able to see it in your statistics. The same pattern can be seen in Search Console, where impressions without clicks have become the norm for many pages.

That’s why AI visibility requires its own metrics: someone needs to ask the questions and read the answers. That’s the whole reason the category exists.

What seven requirements should you set for a tool?

The order is ours. The first three are about ensuring the measurement is reliable; the last four are about ensuring it delivers results.

#	Requirements	How to check this in five minutes
1	Does it measure the engines your customers use?	Ask for the list. Is Google AI Overviews on it?
2	Does it repeat the measurement under the same conditions?	Ask how often each question is asked — and by whom
3	Does it distinguish between ‘cited’ and ‘recommended’?	Ask to see the two figures separately
4	Does it correct anything — or does it just report?	Ask for an example of a correction that has gone live
5	Do the corrections end up in your own system?	Ask what happens if you cancel the service tomorrow
6	Is the price publicly available?	Find the pricing page. If there isn’t one, you won’t be able to set a budget
7	Does it label AI-generated content?	Ask to see the labelling on a finished article

1–3: Does it measure anything consistently at all?

The engine list is the easiest thing to check and the easiest to overlook. Many tools measure ChatGPT and Perplexity because they’re the cheapest to call, and skip over Google AI Overviews — even though that’s the page your customers see without having selected a chatbot. We measure ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews ourselves, and we don’t mention engines we don’t actually measure.

Repetition is more important than it sounds. AI responses aren’t deterministic: the same question can yield different answers from day to day; the models are constantly updated; and your own conversation history influences what you see in the chat. A single response is therefore a sample, not a measurement — the difference lies in the repetition. A tool that shows you a single response and calls it a score is, above all, measuring randomness.

The third requirement is the one the field most often confuses. Being cited as a source and being recommended as a company are two different outcomes, each with its own criteria. We asked four AI engines 576 customer questions about 465 Danish companies: the technical setup was a strong predictor of whether a website was used as a source (p=0.000) — and had no bearing on whether the company was recommended by name (p=0.988). The full analysis can be found in ‘Cited or Recommended’. A tool that combines these two figures into a single percentage hides precisely the difference you need to prioritise.

4–5: Are any changes being made — and will you stick with them?

Your own website is the most important platform, and it’s also the one you can do something about yourself. In our analysis of 576 AI responses, 4,034 sources were cited across 494 domains, and 61 per cent of these were the companies’ own websites — strikingly close to the 57 per cent found by Profound in its much larger dataset. Home turf carries significant weight in both surveys.

At the same time, the gaps on these sites are trivial. When we scanned 465 Danish company websites visible in search results in July 2026, 89 per cent were missing an llms.txt. It’s not difficult work — it’s work that nobody has got round to doing. That’s why question four is the most costly to get wrong: if you only buy the survey, you’re buying a reminder.

The fifth requirement is the one nobody thinks about until they cancel their subscription. The corrections should be incorporated into your own system — your CMS, your files, your structured data — so that they remain in place the day you cancel your subscription. With us, that’s how it works: you lose access to new corrections and to the monitoring service, not the work that’s already been done. Ask your competitors the same question, and ask for their answer in writing.

6–7: Price and labelling

A publicly available price is a requirement, not a mere detail. Several tools in this category do not disclose a price until a sales meeting, and for a company that needs to draw up a budget for the next quarter, this is effectively a ‘no’. Furthermore, the price should be based on something tangible — the number of websites and the number of monitored queries — not on arbitrary feature flags.

Labelling AI-generated content is a new development and is rapidly becoming a standard requirement. Article 50 of the EU’s AI Regulation requires synthetically generated content to be labelled, and the rule comes into force on 2 August 2026. If the tool produces text for your website, the labelling should be applied automatically — and a human should approve it before anything is published. With us, both features are built in: drafts are marked as AI-generated, and your approval is what sends them on.

How much does an AI visibility tool cost?

Prices range from free plans to just over 10,000 kr a month for enterprise packages, which a small business rarely needs. A full breakdown of the providers’ own prices can be found in ‘How much does it cost to be seen in AI?’. By way of comparison, our own pricing structure looks like this:

Plan	Price/month	Websites	Monitored AI questions	Is this right for you?
Free	0 kr	1	—	No — only GEO Score
Starter	299 kr	1	25	Yes — Auto Fix
Pro	799 kr	3	100	Yes + content rewrite
Business	1,499 kr	10	300	Yes + team, API and all integrations

All prices are exclusive of VAT. The ‘Free’ version requires no subscription and only provides the GEO Score. There is a 14-day free trial without a payment card for the paid plans, and annual payment saves you two months’ worth of fees. Full details can be found on the pricing page.

What can no tool promise you?

That AI will start recommending your business. That goes for us too. The technology is the entry ticket: it’s a strong predictor of whether you can be used as a source, but says nothing about whether you’ll be recommended by name — industry, reputation and demand carry more weight in that regard. If a tool promises you a ranking in ChatGPT, it is promising something it has no control over.

The closest research comes to a promise is still just a promise, not a guarantee. The up to 40 per cent from the Princeton study mentioned above is an effect on content visibility — not a reserved spot in the search results, and not a figure anyone can guarantee you.

How do you test a tool in an afternoon?

Five steps, in this order:

1. Write down five questions that your customers actually ask — not your company name, but the problem they’re facing. Here are ten examples to get you started.
2. Ask them yourself in ChatGPT and Perplexity, and note down who is mentioned. It takes twenty minutes and gives you a benchmark.
3. Run the free assessment using two or three tools on the same website on the same day.
4. Compare the responses to requirements 3 and 4: can you see ‘quoted’ and ‘recommended’ listed separately — and is there a button that fixes anything?
5. Repeat in a week’s time. A tool whose figures fluctuate wildly without anything having changed on your site is measuring noise.

If you’re unsure what to look for in your own score, check out the guide on ‘What is an AI visibility score?’.

Frequently asked questions

What is an AI visibility tool?
A tool that asks AI engines the questions your customers ask, and measures whether your business is mentioned and your website used as a source. The best ones also fix the technical aspects of your site — llms.txt, structured data, robots.txt — so that the machines can read your content.

Can’t I just ask ChatGPT if it knows my business?
You can, but it’s a random sample. The answers aren’t deterministic, and your own conversation history influences what you’re shown. If you want to see a trend, you need to ask the same question in the same way over and over again — that’s the whole difference between an impression and a measurement.

How much does an AI visibility tool cost?
From free plans to just over 10,000 kr a month at the higher end. Our own plans start at 299 kr/month for one website and 25 monitored questions, and go up to 1,499 kr/month for 10 websites and 300 questions, all excluding VAT and with a 14-day free trial with no credit card required.

Does it require technical knowledge?
Not with a tool that carries out the corrections for you. You should only need to approve the parts that require your business knowledge — what your company is called, what you do, and who your target audience is. If the tool requires you to edit HTML yourself, you’re effectively back to a to-do list.

Which AI engines should a tool measure?
The ones your customers use. We measure ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews. Take particular note of the last one: it reaches people who haven’t selected a chatbot, and it’s often omitted because it’s more expensive to measure.

Can a tool guarantee that I’ll be mentioned in ChatGPT?
No, and it’s a red flag if anyone promises that. In our own data, technical setup was a strong predictor of whether a website was cited as a source (p=0.000), but had no bearing on whether the company was recommended by name (p=0.988). You can measure, correct and monitor progress — but you cannot buy a ranking.

What happens to the corrections if I cancel my subscription?
With us, they remain in place because they are entered into your own system. You’ll lose access to new corrections and to the monitoring service. It’s worth getting this confirmed in writing with any tool you’re considering — the models vary.