Best tool to track AI search mentions: coverage is published per tier, not per vendor, and at five of nine vendors the surfaces move when you change plan

The engine count on a vendor's website is a fact about one tier, and five of the nine vendors we read change it when you change plan. Scrunch publishes "4 LLMs supported" on Core and "9 LLMs supported" on Enterprise. Profound's free trial covers "ChatGPT, Gemini, Google AI Overviews" and its Enterprise tier publishes "Up to 9 Answer Engines tracked". Peec publishes "Choose 3 models" on Starter, Pro and Advanced alike, and "Up to 13 LLM models tracked" at Enterprise. Depra covers four engines on every plan and offers the rest as Enterprise add-ons. And we move too: our engine row reads five, with Claude as a per-plan yes or no. So the first question to ask any vendor is not how many surfaces it reads, it is how many the plan you are about to buy reads.

The two vendors that publish one engine list for every paid tier

This is worth knowing because it is rare. Rankscale lists the same engines on each of its published plan cards from Pro upward, and publishes a per-engine credit weight instead of a per-tier engine count, so what changes between its tiers is how much you can run rather than what you can run. Ahrefs Brand Radar names one platform set and meters it with one rule: "Each prompt uses 1 check per platform, location and update." Both of those are easier to buy from, because the thing you compared is the thing you get.

Otterly.AI sits in between and is clearer than either about what it is doing. Its base four, "ChatGPT, Google AI Overviews, Perplexity, MS Copilot", is identical on Lite, Standard and Premium, and "Claude, Google AI-Mode, Gemini as extra Add-ons" is also identical on all three. So its surface set does not move with the tier, it moves with a separate purchase that any tier can make.

How the nine sort on this one criterion

Vendor	Does the surface set move with the tier?	What changes, in the vendor's own words
Peec AI	Yes, most of all	"Choose 3 models" on Starter, Pro and Advanced; "Choose from all models" and "Up to 13 LLM models tracked" at Enterprise
Profound	Yes	Trial: "ChatGPT, Gemini, Google AI Overviews". Enterprise: "Up to 9 Answer Engines tracked", with a capability list naming Perplexity, AI Mode, Copilot, DeepSeek, Claude and Exa Search
Scrunch	Yes	Core: "4 LLMs supported". Enterprise: "Expanded model coverage" and "9 LLMs supported"
Depra AI	Yes, at the top only	Four engines on every plan; Enterprise adds "Every other model as an add-on: Microsoft Copilot, Claude, Google AI Mode and more"
AI Knows Us	Yes, on one surface	Five apps, with a per-plan "Claude included" row that reads No on three of our plans
Semrush	Qualitatively	Its FAQ names six tracked surfaces, then "Enterprise AIO adds additional depth and coverage for multi-market programs"
Otterly.AI	No, it moves by add-on	The same base four and the same three add-ons on every published tier
Rankscale	No, above its entry tier	The same engine list on each published plan card from Pro up, with per-engine credit weights instead
Ahrefs Brand Radar	No	One platform list, one metering rule

A second thing that moves the count, and it is Google

Before comparing any two engine counts, check how each vendor counts Google. Rankscale's plan cards read "Google AI Mode, AI Overviews, Gemini" as three entries. Scrunch's Enterprise list carries "Google AI Mode" and "Google AIO" as two of nine. Otterly has Google AI Overviews in its base four and sells Google AI Mode and Google Gemini as two separate add-ons. Depra counts Google AI Overviews as one of four and offers AI Mode separately. Semrush's FAQ names Google AI Mode and Google AI Overviews as two of its six.

So a vendor counting Google three times and a vendor counting it once can read exactly the same Google surfaces and publish counts two apart. Normalise that before you compare anything, and when you do, ask which of the three each vendor actually reads, because Google documents that AI Mode and AI Overviews "may use different models and techniques".

The measurement problem differs per surface, and the price lists prove it

The clearest evidence that these are not interchangeable rows on a feature grid comes from Rankscale's own credit calculator, which splits its list into GUI engines and API engines. AI Overview and Copilot sit under GUI. Perplexity's Sonar family, several GPT versions, several Gemini versions, Mistral Large and two Claude models sit under API. The weights follow the split: ChatGPT, Gemini, Perplexity and AI Mode each cost 0.25 credits per run, DeepSeek 1, and Claude 2.

Depra publishes the same idea as plumbing rather than price, naming consumer web interfaces for ChatGPT and Gemini, a Google results page for AI Overviews, and Perplexity's Sonar model. And it publishes the consequence: "ChatGPT, Gemini and Google AI Overviews are checked daily. Perplexity is checked weekly on Starter, Growth and Scale and daily on Enterprise." A per-surface cost becomes a per-surface cadence, which an engine count hides completely.

One vendor's own two pages disagree, and that is the limit of this method

Read on the same day, 20 October 2026. Semrush's pricing page renders "Unlimited prompts to track daily" on its Advanced card. Semrush's AI visibility solutions page answers "How many prompts can I track?" with "25 prompts on the AI Visibility Toolkit, 50 on Semrush One Starter, up to 200 on Advanced." Both render. Both are published by the same company about the same tier.

This is the honest ceiling on any audit of published pages, including this one: a buyer who reads one page gets one answer and a buyer who reads the other gets a different one. The fix is to put the number in the email and get it back in writing, naming the tier. Nobody loses an argument about a sentence they can quote.

Where rivals beat us, said plainly

We read five surfaces and six of the vendors here publish daily or faster checking where we check weekly with extra checks on request. Rankscale schedules hourly to monthly. Otterly, Peec, Rankscale and Profound publish unlimited or near-unlimited seats where we publish one full-access login on every plan. Scrunch, Peec, Profound, Otterly and Rankscale all publish single sign-on and we publish none. Several vendors read surfaces we do not read at all, including Microsoft Copilot and Google AI Mode.

What we publish that is harder to find: the collection route for each of our five surfaces, the unit as one buyer question on one app once, the Claude multiplier of 3 with the reason for it, and a rule that returns the credit when a check fails.

So how to choose, in the order that actually works

1. Name the surfaces you need, by asking your five real buying questions on each of them yourself and seeing which ones name anybody in your category. A surface that does not answer commercially in your trade needs no budget.
2. Check each shortlisted vendor's cheapest published tier against that list, not its homepage. Four of the eight common surfaces move by tier somewhere in this market.
3. Convert your own number into the vendor's unit: questions by surfaces by places by weeks. A check, a credit, a response, a prompt slot and a monthly AI answer are five different units on these nine pages.
4. Get the tier limit in writing, because one vendor's own two pages already disagree about one of its tiers.

Counting engines is the first thing buyers try and it settles less than it looks, which is worked through separately in our count of 23 vendor documents. The twelve pricing documents behind the figures here are audited in a companion page on which vendors publish a price you can read, one of 147 live pages of ours in this category with no durable inbound link, counted across 2,050 stored bodies.

Questions about buying an AI search tracking tool

Should I buy the tool with the most engines?

Buy the tool that reads the surfaces your buyers use, at the tier you can afford, on the cadence your decision needs. A ninth surface that your customers never open is worth less than a second market on a surface they do, and at two vendors here the ninth surface arrives only with an enterprise contract.

Why do vendors publish different engine counts for the same engines?

Three reasons, all visible on their pages. They count Google between one and three times. They count model versions rather than surfaces in some cases, which is how a list reaches thirteen. And they are quoting different tiers, because five of nine change the set by tier.

What is the single most useful thing to ask a vendor?

Convert my workload into your unit. Give them your question count, your surface list, your markets and your cadence, and ask what that consumes per month on the named plan. Four of these nine can answer it from their own published page, which is a strong signal about what buying from them will feel like.

Is a free trial enough to judge one?

It is enough to judge coverage and the interface, and not enough to judge a trend, because a trend needs more cycles than a trial has days. Use the trial to check that the surfaces you need are actually readable on the tier you intend to buy, which is the thing most likely to differ from the marketing page.

Do any of these vendors publish no engine list at all?

One of the twelve properties we read names no individual surface on the page we fetched: brandlight.ai, whose visible text ran to roughly twelve thousand characters without naming an engine. That is a reason to ask rather than a verdict on the product, and the question to ask is the same one as above.

The one sentence that separates a shortlist

Ask each vendor: on the plan whose price is on your website, exactly which surfaces will I be able to read, how often, and from how many places. Five of the nine will have to give you a different answer from the one on their homepage, and we are one of the five, because our own Claude row is a yes on some plans and a no on others. The vendor whose answer is identical to its published page is the one you can plan a year around, and that is a more useful thing to buy on than any engine count.