Which AI search tracking tool covers the most engines? 23 vendor documents counted on 20 October 2026, and the count is the wrong question
Eighteen of twenty three responses named all five of ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, so the surface list no longer separates these products. Three of eleven vendors publish how each engine is actually reached, and that is the column that still sorts them.
Published by AI Knows Us · Updated 20 October 2026
On 20 October 2026 we requested 23 documents belonging to eleven companies that sell AI search tracking, including our own, ran no JavaScript, and counted the engine names in the visible first response. Every one of the 23 answered HTTP 200. Eighteen of the 23 named all five of ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. One named none of the five across 1,812 words. So the honest answer to which tool covers the most engines is that the published surface list has stopped being a difference: almost everybody prints the same five names, and the headline counts above those names count different things. The column that still sorts these products is whether the vendor publishes how each engine is reached, and three of the eleven do.
We sell a tracker in this category, so our own pages are in the count and are measured with the same request. Where a rival publishes more than we do, it is named below.
What the 23 responses named
One request per URL, a desktop browser user agent with an Accept-Language header, no cookies, scripts and tags stripped. An engine counts as named when its name appears in that visible text. A response under 400 visible words is recorded as not usable rather than as an absence.
| company | documents read | of the five engines named | count the company states in its own words |
|---|---|---|---|
| Otterly.AI | 3 | 5, 5, 5 | "all 7 major AI search engines" |
| Profound | 3 | 5, 5, 5 | no count in the text we read |
| Rankscale | 2 | 5, 5 | "17+ AI engines" |
| Depra AI | 2 | 5, 5 | four engines, named individually in its method page |
| Athena | 2 | 5, 5 | no count in the text we read |
| Peec AI | 2 | 4, 5 | "Up to 13 LLM models tracked" on the top tier |
| Ahrefs Brand Radar | 1 | 5 | no count in the text we read |
| Conductor | 1 | 5 | no count in the text we read |
| Semrush | 1 | 4 | six named in its own FAQ answer |
| Scrunch | 2 | 2, 4 | no count in the text we read |
| Brandlight | 1 | 0 | no engine named in the response we read |
| AI Knows Us, ours | 3 | 5, 5, 5 | five surfaces, named individually |
Two rows need reading rather than quoting. The Scrunch home page names GPT-4 by model name rather than ChatGPT, so a probe looking for the product name undercounts it, and its pricing document named four. The Brandlight home page describes how a brand appears "across AI" and named no individual engine in the 1,812 words served to us, which is a publishing choice on one document rather than a statement about its coverage.
Three published counts that cannot be compared with each other
Seven, thirteen, seventeen and four all appear above, and a buyer who lines them up is comparing four different nouns.
- Otterly states seven major AI search engines and writes the list out: ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Microsoft Copilot and Claude. Google appears three times in that list as three different products, which is correct, because an AI Overview, AI Mode and Gemini are three different answers to the same question.
- Rankscale states 17+ AI engines and names ChatGPT, Perplexity, Claude, Gemini, Copilot and DeepSeek inside that claim.
- Peec states up to 13 LLM models tracked on its top tier, and models are not surfaces. Several models sit behind one app.
- Depra names four engines and is the smallest published number in the set.
So the comparison worth making is never between two headline numbers. It is between the enumerated list on your own quote, surface by surface, with the cost of any that is extra written next to it. A list can be compared. A count cannot.
The vendor with the smallest engine count publishes the most about each one
Depra AI names four engines and publishes, at a method page dated 19 October 2026 with a named author, how each one is captured: ChatGPT and Gemini from the consumer web interfaces at chatgpt.com and gemini.google.com, Google AI Overviews read from a real Google results page including whether an Overview appeared at all, and Perplexity through its Sonar model with live web search. The same page publishes the cadence per plan, that questions are asked at country level across 40 markets with up to 10 per project, that prompts containing the brand name are excluded from scores by default, and that a mention below 50 percent confidence never counts.
Ahrefs publishes a Brand Radar method post with a named author, a named reviewer and an update date, and it states something a buyer needs: Brand Radar models prompts from recurring high demand topics in a keyword database rather than putting one buyer's question to each app. That is a different instrument from a prompt tracker, it is published plainly, and knowing it changes which product answers your question.
Our own methodology page names the route to each of the five: ChatGPT and Gemini through a third party data provider, Perplexity through its official Agent API with live web search, Claude through the official Anthropic API with live web search, and AI Overviews through a licensed search results provider. On the count of named engines whose collection route is published, we are first in this set at five of five, Depra is second at four of four, and the other nine companies published no per engine route in the documents we read.
Where this ranking stops flattering us
Three things in that paragraph deserve the other half said out loud.
- Our product reads five surfaces and our published datasets cover two. Both of our large audits, 597 readings in October and 1,260 readings on a wider question set, were run on ChatGPT and Gemini. Perplexity, Claude and AI Overviews were in neither. Five is what the product reads and two is what we have published numbers from, and those stay separate in every sentence.
- Depra's method page is dated more recently than ours and names its author. It also publishes a 12 tool comparison on engines and price, which this site deliberately refuses to publish for anybody.
- Our own citation record in this category is poor and published. Six readings of 1,260 cited our domain, zero of those in the United States, and zero on Gemini in 921 readings across both audits.
The four questions to put on the quote instead of counting engines
- Write the engines out individually and price each one. Two quotes can be compared on a list and never on a phrase.
- Ask how each one is reached. A consumer app, an official API and a licensed search feed are three different instruments reading the same brand, and they disagree for reasons unrelated to your marketing.
- Ask whether a Google surface counts once or three times. AI Overviews, AI Mode and Gemini are three products and one vendor's "Google" may mean any of them.
- Ask for the raw answers. Coverage is a claim about inputs. The answers on disk are what lets somebody else check the number you report.
If you are still working out which of the four product categories sold under this phrase you are shopping in, the split is set out in what each category of AI visibility tracking tool actually covers, and the answer changes which of the counts above matters to you at all.
Questions buyers ask about engine coverage in AI visibility tools
Which tool reads the most AI engines?
On published claims read on 20 October 2026, Rankscale states the largest number at 17+ AI engines and Otterly states seven with the full list written out. Those two numbers are not measuring the same object, and 18 of the 23 documents we read named the same five surfaces, so the published list is not where these products differ any more. Compare the enumerated list on your quote.
Does more engines mean a better tool?
It means a larger bill and more rows. The engines that matter are the ones your buyers actually open, and for most businesses that is two or three. Buy the breadth you will read every month, then spend the rest on getting the answers exported so somebody can check them.
Why do two tools report different numbers on the same engine?
Because they reach it differently. A consumer web interface, an official API with live search and a licensed search results feed return different answers to the same question on the same day, and the prompt set, the country, the repeat count and the sign in state all move the result as well. Pick one instrument, keep it frozen, and measure your own movement against itself.
Do any of these vendors publish how they collect the data?
Three of the eleven in this set do, read on 20 October 2026: Depra AI at a dated method page naming four engines and the capture route for each, Ahrefs at a Brand Radar method post with a named author and reviewer, and ourselves at a methodology page naming the route to each of five surfaces. The other eight published no per engine collection route in the documents we read, which is a fair question for a demo rather than a verdict about those products.
Is Google AI Overviews one engine or three?
Google serves AI Overviews, AI Mode and Gemini, and they answer the same question differently. Otterly's published list treats them as three entries and that is the version a buyer should hold a vendor to. If a quote says "Google", ask which of the three it means and whether the other two cost extra.
The sentence to take into the demo
Coverage used to be the whole conversation in this category and 18 of 23 documents have now settled it the same way. The five names are nearly uniform and the instruments behind them are not, so the engine count has stopped being the question and the route has become it. Ask for the route, surface by surface, in writing. Three of the eleven companies here have already filled that column in, and the answer separates these products in five minutes.