8 AI Visibility Tools Compared (2026): Features and Prices

Short answer: on 4 August 2026 we opened the pricing pages of eight AI visibility tools. All eight publish an entry price, ranging from $15/month (MaxAEO Starter, billed annually) to $250/month (Scrunch Growth, billed annually). Two of them — Peec AI and Trakkr — render their prices only in the browser, so a fetch returns an empty table where a person sees numbers; we read those two on screen. The tools disagree on what a "prompt" even is, so per-prompt price comparisons are mostly fiction. And none of the pricing pages we read said how many times each prompt is asked, which is the number that decides whether any of it means anything.

On 5 August we went back and compared them on features too — engines covered, refresh cadence, how prompts are supplied, and how many times each prompt is asked. Of the 32 feature cells that describe the eight vendors, 18 are blank, because the vendor does not publish the answer. We have marked those not documented rather than guessing, and the section explains why that is not the same as no.

We build peekr, an AI visibility tool. That is a conflict of interest, so it goes in the second paragraph rather than a footer. We put ourselves in the same tables below, on the same rules, along with the columns where we currently lose.

What is an AI visibility tool?

An AI visibility tool asks AI assistants a fixed set of questions on a schedule, records whether your brand appears in the answers, and reports the result over time. It is a measurement product: it does not put you into AI answers, it tells you whether you are in them. Most tools also grade your pages and suggest fixes, but the grade and the appearance are two different numbers and should be read separately.

Prices we verified on 4 August 2026

How we got these: we fetched each vendor's public pricing page with an automated reader on 4 August 2026 and recorded what was on it. We did not create accounts, and we have not tested any of these products. Everything below is what the vendor published about itself, not a performance judgement.

Where the ladder goes above the entry plan, for the five that publish a full ladder:

• Profound — Growth $399/mo, 100 prompts, "3 Answer Engines tracked"; Enterprise custom, "Up to 9 Answer Engines tracked".
• Otterly.AI — Standard $189/mo, 100 prompts; Premium $489/mo, 400 prompts. Claude, Google AI Mode and Gemini are add-ons from $9–$29/mo.
• Scrunch — Growth $417/mo annual ($500 monthly), 700 custom + 2,500 industry prompts.
• AuraMetrics — Medium $95/mo, 100 prompts, 5 domains, adds Claude and Perplexity; Pro $207/mo, 150 prompts; Enterprise custom, 1,500 prompts per domain.
• MaxAEO — Growth $119/mo annual ($149 monthly), 100 prompts; Pro $319/mo annual ($399 monthly), 400 prompts across 4 brands.

Two caveats that matter more than the numbers. Annual and monthly prices are mixed above because vendors headline different ones — always check which you are reading. And pricing pages change and get A/B tested — last week we read a vendor's homepage and its pricing page on the same day and got two different sets of numbers, and we could not establish which was current. Treat this table as a dated snapshot, not a live feed.

The unit is not the same, so the per-prompt math is fake

A lot of comparison posts divide price by prompt count and rank the result. We have done that arithmetic ourselves. Reading these eight pages back to back convinced us it is misleading:

• Scrunch splits "custom prompts" from "industry prompts" — 350 and 1,000 at Starter. Those are not the same good.
• AuraMetrics meters prompts and analyses per month per domain separately, plus a domain count and a seat count.
• Geneo does not sell prompts at all at the tier we read. It sells 1,000 credits per month, and the page we read does not state what a credit buys.
• Otterly.AI puts three engines behind add-ons, so the sticker price and the price for the engine set you actually want are different numbers.

If you want a comparison that survives contact with an invoice, price the exact configuration you need — engines included, prompts of the kind you'll actually use, seats, domains — and ignore the headline.

What these vendors do better than us

Since we are a competitor writing this, it is worth being specific about where the tools above are ahead, on evidence we can point at:

• Otterly.AI publishes the whole ladder, including the add-ons. Four tiers, monthly and annual, and the price of each extra engine. Most software pricing pages hide the second half of that.
• Profound states its biggest limitation on the price tag. "ChatGPT tracking only" is written on the $99 tier rather than discovered after purchase.
• Scrunch publishes prompt counts at every published tier, and separates custom from industry prompts instead of quoting one blended number.
• AuraMetrics publishes its prioritisation formula. On their content-optimisation module page, read 29 July 2026, they state the exact weighting used to rank suggested fixes — estimated impact 40%, inverse effort 25%, citation-probability correlation 20%, competitive context 15% — and their citation-patterns module shows a confidence level with a stated cross-validation accuracy range. We have no equivalent formula and no confidence display on our screens. They also sell a $5 / 5-day trial with no auto-renewal, which is a more honest shape than a free trial that bills you silently.
• MaxAEO shows annual and monthly prices side by side rather than headlining the annual number alone.

We have one methodological disagreement with the category, described below: several tools predict how many points a fix will gain you, and we could not find one — ourselves included — that publishes whether those predictions came true. That is a disagreement about method, not a claim that these are worse products. On engine count, prompt volume and refresh cadence they are all ahead of us today.

Features compared: what each vendor's own pages state

Read this line before you read the table. Every cell below is what the vendor's page says about itself, copied on 5 August 2026. It is not a judgement about whether a feature works, and it is not a judgement about whether a feature exists. We have not used any of these products, so we are not in a position to tell you something is missing — only that we could not find it published. Where a page is silent the cell reads not documented, which is a different statement from no.

That distinction is the only thing that makes this table different from the ones ranking above it. Most comparison posts in this category present a feature grid of ticks and crosses without saying whether anyone opened an account. A cross in those grids and a blank in ours mean very different things.

Two rows are weaker than the rest. Trakkr and Geneo render their pricing in the browser, so an automated fetch returns navigation and no table. Their cells carry what we read on screen on 4 August 2026, marked below, and we did not re-verify them on the 5th.

Engines, refresh cadence, and repetition

One conflict we could not resolve: on 4 August our on-screen read of Peec AI's Starter tier recorded a choice of three models, while the page text we fetched on 5 August lists six models without a tier qualifier. We do not know which is right and have left both readings visible rather than pick one.

The emptiest column is the one that decides whether any of the other numbers mean anything. Eight vendors, eight blanks. Cadence is well covered — four of the eight state "daily" on every tier — but cadence is how often a measurement is taken, not how many answers each measurement is built from.

Profound is the near-exception, and only by arithmetic. Its page publishes a monthly response count beside the prompt count: "50 unique prompts" and "1,500 responses monthly" at Starter, which is ChatGPT only; "100 unique prompts" and "9,000 responses monthly" at Growth, which is "3 Answer Engines tracked". 50 × 1 engine × 30 days = 1,500. 100 × 3 engines × 30 days = 9,000. Both identities hold exactly, which is what you would see if each prompt were asked once per engine per day.

Profound does not say that, we did not test it, and other arrangements produce the same totals. We are pointing at arithmetic, not describing their product. What is not in dispute is that they publish a response count at all — no one else in the table does, and without that number you cannot even attempt the check.

Prompt supply, and what each page leads with

This axis is easy to skip and it changes what a score means. If the answers you are graded on were generated once for your whole industry and then scanned for everyone's brand, that is cheaper to run and it gives you a peer comparison for free. If they were generated from your questions, they are about you and nobody else. Neither is better in the abstract — but only two of eight pages tell you which one you are buying. Outside this table it is clearly possible to disclose: Ahrefs' Brand Radar states plainly that its database is "powered by search-backed prompts, not synthetic ones" and publishes the volume behind it.

Counting the blanks. Across the four columns that describe vendor facts — engines, cadence, repetition, prompt supply — the eight vendors give us 32 cells. 18 of them are not documented. Engine coverage is published by all eight; cadence by four; prompt supply by two; repetition by none.

What no pricing page told us: how many times each prompt is asked

That empty column deserves its own section, because it is the number the whole product category rests on. Not one of the pages we read stated how many times a prompt is asked per cycle. Several state a cadence — "daily tracking" — which is a different thing. Cadence is how often you take a measurement. Samples is how many answers each measurement is built from. A daily measurement built from one answer is a daily anecdote.

This matters because AI answers are not stable. In published work on measuring LLM outputs, a per-cell 95% confidence interval at three samples is about ±36.9 percentage points. That figure is from other people's research, not ours. At one sample there is effectively no interval at all — you have an anecdote with a percent sign on it.

So a dashboard that says "you appear 33% of the time" can mean almost anything unless you know the denominator and the repetition count. Ask any vendor, including us:

1. How many times is each prompt asked per cycle? If the answer is one, the percentage is decorative.
2. Is web search on? With grounding off you are measuring the model's memory, not AI search. We measured this on ourselves: 340 stored responses, zero citations, because our own connectors were not grounded at the time. Turning grounding on raised our cost per response roughly 11× in our own measurements. That cost is why it is often quietly off.
3. Does the visibility score include actual citations, or only on-page signals? If it is only on-page signals, the score can improve while nothing changes in any AI answer.
4. Are the raw answers kept? If not, you cannot re-check a number later, and neither can they.

We do not know how these eight vendors answer questions 1 and 2, because we did not test their products. We are telling you what to ask, not what they would say.

Every "best AI visibility tools" list is written by one of the tools

On 28 July 2026 we opened the pages ranking for the obvious buying query in this category and classified who wrote them. Of the six pages we read, one was independent editorial — a Zapier review whose author described creating 20+ accounts and booking walkthroughs to test the products. The other five had an interest in their own ranking: an agency list that included the agency, a vendor's own list with the vendor at the top, small vendors publishing "A vs B" pages about companies they compete with, and a crowd-sourced directory.

None of that is dishonest by itself, and one date and one query is a thin sample. It is a description of how a young category writes about itself when there is not yet much independent coverage to link to.

Then, on 4 August 2026, we ran a backlink link-intersection query across a set of competitor domains — the domains that link to the most of them at once. The top of that list was other AEO/GEO tools: trakkr.ai, scrunch.com, geneo.app, maxaeo.ai. That was a single run against one vendor's backlink index, so treat it as a hint rather than a finding; we are re-running it later.

But the direction is consistent with what we saw by hand. In this category, the links flow through the vendors' comparison posts about each other. It is a room where the reviewers are the reviewed.

We are now in that room too. This post is a vendor writing about competitors. You should discount it accordingly, and the specific way to discount it is: trust the price table (it is copied off their pages and dated) and be sceptical of any sentence where we characterise a competitor's quality. We have tried not to write those sentences.

Our own data says publishing this post is bad for us

Here is the awkward part. We measure which pages AI answers cite, and vendor listicles do badly for the vendor who wrote them.

On the Google AI Overview surface, in our own measurements: listicle-type pages were the largest single citation channel — 38% of 154 citations. But when one of those listicles got cited, the publisher's own name survived into the answer only 22% of the time (n=21). Product pages, cited on the same surface, carried the publisher's name 89% of the time. Writing the list opens the door; it mostly lets someone else walk through it.

That 89% probably flatters product pages, by the way — a product page may get cited because the model already intended to name the brand, in which case we have the arrow backwards. We haven't separated those two, so read it as a gap worth noticing, not a measured effect.

We also tried ranking ourselves. In a separate run (n=21) we put ourselves first in our own list, the list got cited, and the answer named a different company first. Self-assigned rank is not adopted.

All of the above is one surface (Google AI Overview / Gemini) and small samples from one brand. We have no equivalent data for ChatGPT — we have not confirmed what its retrieval surface even is — and zero measurements for Perplexity. We are not going to pretend otherwise.

So by our own numbers, this post is more likely to help the companies in the table than to help us. We published it because we would rather be the source of a dated, checkable price table than win an argument, and because the alternative — writing nothing while eight other vendors write about each other — has not been working either.

Where peekr falls short today

We would rather you learn this here than after signing up.

The measurement that made us build this

On 29 July 2026 we ran our own product against our own site. On-page GEO readiness came back A / 90. AI exposure came back 0 / 2 — that is one question, two engines, one sample each, web search off. The names that came up instead of ours were Profound and Otterly.AI, competitors in the table above.

Two honest qualifications, because this number gets over-quoted:

• The A was not wrong and was not supposed to produce exposure on its own. Our sitemap had been submitted that morning; Google index count was 0. A page that isn't indexed cannot be cited no matter how well built it is.
• The 0/2 is no longer a usable baseline. It was measured with web search off. Our connectors now run grounded, and comparing a grounded number to an ungrounded one is exactly the mistake we tell other people not to make. We have to re-measure zero before we can claim anything moved.

More recent, and blunter. As of 4 August 2026: DataForSEO's backlink index has no row for our domain at all — not "zero backlinks", but "this vendor has never seen this domain". And Bing Webmaster Tools reports our homepage, pricing page and blog index identically — "Discovered but not crawled — URL cannot appear on Bing", discovered 29 July and 2 August. Bingbot is allowed in robots.txt and gets a 200 in 0.2 seconds and 36KB when it asks. It just isn't coming back. Bing's own interface states that "SEO/GEO processing can be done only on indexed URLs."

Three independent measurements saying the same thing: nobody is reading us yet. Our on-page grade is an A. It is doing nothing.

So how should you choose?

Of the four questions below, only the first can be answered from the vendors' own pages. That is what the feature tables above show: engine coverage is published by all eight and refresh cadence by four, and after that every page goes quiet. So the rest of this list is questions to put to a salesperson, not columns you can look up.

1. Engines you actually care about, at the tier you'll actually buy. Engine coverage is the most common paywall in the table above — Profound's $99 tier is ChatGPT only, Otterly puts three engines behind add-ons, AuraMetrics adds Claude and Perplexity at $95.
2. Samples per prompt, and grounding on/off. See the four questions above. This determines whether the percentages are real.
3. Whether the "score" contains real citations or only on-page signals. A score built purely from on-page signals will move when you edit your HTML, whether or not any AI answer changes. Our own A/90 vs 0/2 is what that looks like from the inside.
4. Whether anyone checks their own predictions. Several tools in this space predict how much a fix will improve you. We could not find a vendor that publishes whether those predictions came true — including us. This is the part we are building toward, and we are not there.

We are not going to tell you peekr is the right answer today. On engine count, prompt count and refresh cadence, the products above are ahead of us, and we just listed the gaps. What we can offer right now is a free page-level GEO audit that calls no AI engine, needs no signup and stores no results, plus a single snapshot of what two engines say about you — with web search on by default, and labelled as a single snapshot, because that is what it is.

Run the free audit on one URL No signup, no engine calls, nothing stored. It reads the page and shows what a retrieval engine can and cannot see on it.

FAQ

How much do AI visibility tools cost?

Between $15 and $500 per month for the entry and mid tiers we verified on 4 August 2026, with enterprise tiers quoted on request. The spread comes mostly from prompt volume and engine coverage, not from feature depth.

Which AI visibility tool tracks the most engines?

Of the pages we read, Scrunch listed the widest set at its published tiers (ChatGPT, Claude, Gemini, Perplexity, Google AI Mode and AI Overviews, and Meta), and MaxAEO advertised eight platforms with all eight at Enterprise. Profound reaches "up to 9 Answer Engines" at Enterprise but sells ChatGPT alone at its entry tier. We did not test any of these products; this is what their pages say.

How often do AI visibility tools refresh their data?

Four of the eight we checked on 5 August 2026 state "daily" on every published tier — Otterly.AI, Profound, Peec AI and MaxAEO. The other four (AuraMetrics, Scrunch, Geneo, Trakkr) do not state a cadence on the pages we could read; AuraMetrics meters credits per module run instead of publishing a schedule. Cadence is not the same as sample size, and it is the weaker of the two numbers.

Do any of these tools say how many times they ask each prompt?

None of the eight states it. Profound is the only one that publishes enough to derive a figure: its response counts per month (1,500 at 50 prompts and one engine; 9,000 at 100 prompts and three engines) are exactly prompts × engines × 30 days, which is consistent with one ask per prompt per engine per day. That is our arithmetic on their published numbers, not their claim, and we have not tested it. Ask any vendor directly, including us.

Do these tools get you into AI answers?

No. They measure. Any tool promising placement in AI answers is describing something it cannot control — we don't make that claim and you should be wary of anyone who does.

Why is a vendor publishing a comparison of its competitors?

Because the alternative in this category is a room where every list is written by someone in it, with no disclosure. We would rather disclose, date every number, and say which columns we lose.

Related: why a brand doesn't show up in AI answers · how to track brand mentions in AI search

Prices recorded 4 August 2026 and feature statements recorded 5 August 2026, from public pages, by automated fetch, without accounts. Trakkr and Geneo render in the browser and were read on screen on 4 August; their rows are marked. We have not used any of these products, so every cell is a record of what a vendor published about itself, not a test result. If we have something wrong, tell us and we will correct it with the date of the correction.