Guides · 2026-08-28 · By VSNARY | Emmanuel Orta · 5 views

AI visibility tools compared: what each one actually measures

Every tool in this category runs the same loop. The comparisons all sort on price and engine coverage. The difference that decides whether a tool can help you is which half of the problem it can see.

The category is barely two years old and already crowded. Profound, Peec AI, Otterly, Scrunch, AthenaHQ, Semrush's AI Toolkit, Ahrefs Brand Radar, and a long tail of free graders. Every published comparison ranks them on price, prompt volume and how many engines they cover.

Those are real differences. They are also the small ones. Nearly every tool in this list runs the identical loop: take a list of prompts someone wrote, send them to model APIs on a schedule, count how often your brand is named, and report a percentage over time. The vendors themselves are fairly open about this — the honest comparisons in the category say plainly that these tools measure the gap and do not close it.

So the useful way to compare them is not by price. It is by which half of the problem the tool can observe.

Disclosure: CrawlCheck publishes this comparison and is one of the products in it. Every figure about another company comes from that company’s own pricing or documentation page, linked where it appears. Where CrawlCheck does less than a competitor, the page says so.

The two halves #

A demand-side tool tells you that you are not being cited. It cannot tell you that the reason is a cached challenge page sitting on your robots.txt, because it never fetched your robots.txt. It asked a model a question and wrote down the answer.

The categories, by what each can see #

Prices, side by side #

Published entry prices on 3 October 2026, each from the vendor’s own page: Otterly.AI Lite $29/month (15 prompts), Rankscale Essentials $17/month billed yearly and Pro $99, Knowatoa Starter $59, Semrush AI Visibility Toolkit $99, Ahrefs Brand Radar custom prompts from $50 and the AI Visibility Index $199, AthenaHQ Starter $295 (free tier with 300 credits), Scrunch Starter $300; Profound and Peec AI publish no monthly price. CrawlCheck scans are free and monitoring is $29/month for up to three domains. The full table, with engines covered and a link to every source, is in best AI visibility tools compared.

The label problem, in three parts #

Worth understanding before you buy anything in the first two rows, because it applies to all of them equally and none of it is fraud.

The surface is not the product. The OpenAI API is not ChatGPT — the app performs retrieval, calls tools, and carries personalisation and memory the raw API does not reproduce. Google's AI Overviews is a different system again from the Gemini API. There is no public API for the consumer surfaces, so every vendor measures the thing they can reach and labels it as the thing you asked about.

The sample is somebody's guess. Nobody holds the distribution of what buyers actually type into an assistant — not the vendors, not the model providers. A share-of-voice percentage is a statistic over a hand-picked prompt list presented as a property of your brand. Change the list, change the number.

One answer is one draw. Ask the same model the same question twice and the answers can differ. Ask next week, after a model update, and they differ more. The churn between dated repetitions is the actual signal, and it is the part most tools do not surface, because it makes the headline number look unsteady.

What happened when we scanned the vendors #

Thirty-three SEO and AI-visibility vendors were run through the same supply-side scanner they would point at you — crawler-identity checks, machine files, cache behaviour, render path. None was told in advance; every one got the same checks, from the same client, on the same day.

Several were serving a cached copy of a machine file that no longer matched their own origin — meaning a crawler and a human are told two different things, by a company that sells judgement about exactly that. One could not be measured at all: every path requested, including one that cannot exist, answered identically. That is a firewall speaking, not a website.

The point is not that these vendors are bad. It is that these checks are not niche requirements only a specialist can meet — and the people selling the advice mostly have not implemented it.

Four questions that separate a measurement from a label #

Ask any vendor in this category, including this one:

1. Which surface did you query — the consumer app or an API? If it is an API, the number is about the API.
2. Which model, which mode, which date? Grounded retrieval and generation from weights are different systems and should never be averaged together.
3. Who wrote the prompt list, and what happens to the score if I replace it?
4. What would have to be true for this number to be wrong, and how would I find out? If there is no answer to the fourth, it is not a measurement. A percentage you cannot check is a claim about something nobody measured.

How to actually pick #

If nobody has ever verified that answer engines can fetch your site, start there, because a demand-side subscription measures an outcome you have not yet made possible. Across the public dataset here, 17.4% were serving a stale cached object to crawlers on the bare URL and 5.8% were refusing at least one answer engine outright while serving browsers normally. No prompt monitor can see either.

Once that is clean and stays clean, a prompt monitor tells you something the supply side cannot: whether any of it produced a citation. The two are complements, and the order is not interchangeable.

On the vendor scan above. No scanned vendor domain is published; the 33-vendor figures are aggregates for that reason. CrawlCheck also cannot grade its own site, so check our files with any client you like.

Where CrawlCheck is weaker #

CrawlCheck sits in the supply-side row. Its own answer-engine check is narrow next to the prompt monitors: a small prompt set per monitored site on a single engine, no share-of-voice dashboard against named competitors and no sentiment scoring. If those matter most, a prompt monitor is the better buy; run the free supply-side scan first so its numbers are not limited by access.

Findings behind this #

Every figure above comes from a measurement written up in full, with the date, the method and the raw numbers.

• 33 AI visibility tools measured: a third publish llms.txt, none an entity map
• AI visibility scores measure an API, not what anyone sees in ChatGPT
• Best AI visibility tools compared: prices and engines, with sources
• AI visibility tools that grade you 100% and still sell you the fix

Questions this post answers

What is the difference between AI visibility tools?

Less than the comparisons suggest. Almost all of them run the same loop: a prompt list is sent to model APIs on a schedule and brand mentions are counted. They differ mainly on price, prompt volume and engine coverage. The larger difference is category: prompt monitors measure whether you are named, while machine-layer scanners measure whether the retrieval layer can read you at all.

Do AI visibility tools measure ChatGPT?

They measure model APIs. There is no public API for the consumer apps, so every vendor queries what it can reach. ChatGPT the product performs retrieval, calls tools and carries personalisation that the raw API does not reproduce, so an API result is a proxy, not the surface a customer sees.

Which AI visibility tool should I start with?

If nobody has verified that answer engines can fetch your site, start with a supply-side check, because a prompt monitor measures an outcome that is not yet possible. In this dataset, a stale cached object served to crawlers is the most common finding by a wide margin, and refusing at least one answer engine while serving browsers normally is far rarer; both shares are published live on the dataset page.

Related findings

Comments

Comments are read before they appear. Nothing is published automatically, and no account is needed.

Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.

All guides · The dataset · How the dataset works