The setup is documented. The timing is not.

The category is unusually good at being readable and unusually bad at being accountable.

The good news first. These vendors have gone further than most in publishing documentation for machines, and 9 of the 11 we probed publish an index of their own documentation at a path an agent can read, listing 817 pages between them. That is a category that has thought about machine readers, and it is more than most software can say.

The bad news is what the index does not contain. 0 of 9 vendors state a time to first report. 3 publish a page that walks through setting the product up, and 1 describe onboarding help. When a buyer asks how long this will take, the answer is a question the vendor has chosen not to publish.

So the number has to come from somewhere else, and that is what this study is for. The measurement you want is not in the documentation, it is in the contract, and the questions to ask are below.

There is a second thing in the data that is worth a buyer knowing, and it is not a gap but a shape. Across the nine indexes, 438 pages are interface reference and 379 are product documentation. For several vendors the documentation a buyer would read to set the product up is a minority of what is published, and the rest is how to call the API.

Last verified 2026-10-23. Access rules and documentation change without notice. Re-read the table before you rely on a row in it.

What each vendor publishes, and what we could read

Eleven vendors from the category, probed on 23 October 2026. The rules column is whether we could read the host's robots file, because a rules file behind a bot wall is a different thing from a rules file that is readable.

How we got to a documentation index for each vendor

Three of the nine indexes came from a host whose rules we could not read, which is a finding in its own right and is drawn here rather than hidden in a note.

The red segment is the awkward one. Those vendors serve an open index to an agent while refusing to serve the rules that govern it to the same agent.

How much each vendor publishes, and how much of it is for you

The number of pages in each index, against the number of those pages that are product documentation rather than interface reference.

A large index is not a large guide. The gap between the two bars is the part of the documentation that tells you how to call the product rather than how to run it.

The category publishes for agents and keeps the effort to itself

There is a particular kind of company that appears when you look at the documentation of an AI visibility tool, and it is worth naming because it changes what you should expect from the price page.

A company that sells monitoring and reporting for AI search has an unusual property: it wants to be found by the systems it monitors. It publishes a machine-readable index of its own documentation, it serves that index without a bot wall, and it names the engines and the models it tracks. That is a company that has thought carefully about being quotable.

The same company, asked to say how long its own setup takes, publishes nothing. Not one of the 9 indexes carries a duration attached to a first result, and 3 of them do not even promise a first result in words. There is no contradiction here. The two behaviours come from the same strategy: being findable is valuable and being accountable is a cost, and the category has chosen the first and not the second.

For a buyer that is a predictable and manageable situation. You will not be able to compare vendors on time to value, because none of them publish it. You will be able to compare them on how much they publish for you, and the answer is that most of it is for developers.

What the indexes actually answer

Each field is credited when a vendor's index lists a page whose title or description answers that question. The count is out of the 9 vendors that publish an index at all.

Four things the indexes say about the category

The findings, in the order they matter to a buyer considering a switch.

Three vendors promise a first result. None of them says when.

This is the whole finding in three sentences, taken from the indexes.

Hikoo's quickstart page says: set up your Hikoo account and see your first AI-visibility snapshot. AthenaHQ's getting started page says: set up AthenaHQ and start tracking your brand across AI search. Peec AI's quickstart guide says: the fastest way to understand your brand's AI search presence.

Read them again for the thing that is missing. Each one promises that you will get something to look at, and each one stops before the word that would let you plan. There is no day, no week and no hour in any of them, and no other page in any of the nine indexes fills the gap.

That is not a criticism of the writers. A snapshot arrives when the data has been collected, and the collection is the vendor's job, so a published time is a promise the vendor may not want to make. It does mean that if the time to first result is the thing you are being sold, you have to ask for it, and you have to put the answer in the contract.

Each vendor, with what it published

The nine vendors that publish an index, one at a time, with the size of what they published and which of the six fields it answers. This is the part you would otherwise have to assemble by hand, which is why the folk claim that this category is transparent has survived so long.

Five questions to ask before you buy

The gap is the vendor's to close, and these are the questions that close it.

What planning a rollout on this documentation looks like

The abstract version of this study is easy to nod along to and then get wrong. Here is the sequence, for a team switching a tool in this quarter.

You have read the indexes. You know that one vendor publishes a quickstart, that another publishes a page for adding a website through an onboarding wizard, and that the rest publish API documentation. You also know that none of them will tell you how long the first report takes, so you have to assume a number and make it the vendor's problem.

So you go into the evaluation with a stated assumption rather than an open question. Say two weeks to first signal for the tool with a real setup path, and four for the two whose documentation is mostly about calling an API, because there the setup is the integration. Write those numbers into the evaluation sheet, so that they are visible, and so that the vendor's answer can be compared against something.

Then ask the five questions, and treat the answers as part of the product. A vendor that will put a time on it in the contract is telling you something about how it will handle the other surprise. A vendor that will not is telling you something too, and you can plan around it: you build the internal dashboard yourself, and you budget for the month where the report is late and there is no explanation to point at.

That is what a documentation census is for. It does not tell you the answer. It tells you that the answer is not published, which is the thing you cannot get from a pricing page.

An agent index is not a transparency policy

It would be easy to read this census as a ranking of how open these vendors are. It is not, and the reason is worth stating.

6 of the 9 agent indexes come from a host whose robots file we could read. 3 come from a host that answers an automated client with a challenge page instead of a rules file. For those vendors the index is served openly while the rules governing it are not.

That is not a hypocrisy finding, it is a measurement of what publishing an index means. A machine-readable index is a convenience for an agent that already knows where to look. It says nothing about whether the vendor will answer a question, or how long the answer will take, and it is not a commitment to keep it current.

The one thing it does tell you is which way these vendors think about machine readers, and that is worth something. It is just not the thing you were about to buy.

The five questions this census keeps getting asked

Four readers, and what each of them should take

The same measurement reads differently depending on what you are doing with it.

You are switching a tool in. The cost of the switch is not the licence, it is the weeks before the tool can tell you anything, and this census tells you which vendors will help you find that number and which will not.

You are selling one of these tools. The fastest fix is to publish a time, even a range, on the page a buyer lands on. Nine competitors already publish a machine-readable index of their documentation, and not one of them publishes the number a buyer is actually shopping for.

You are writing a comparison of this category. The documentation is the easiest axis to compare on and the least useful, because it measures the size of the marketing effort rather than the quality of the product.

You are building on top of one of these tools. The interface pages are where the useful detail is, and they are more or less complete depending on the vendor rather than on the plan.

What this study does not tell you

Where this census sits in the rest of our work

Three studies that bear on the same question: what a vendor's own page lets you find out before you talk to them.

How we checked

We probed 11 vendors from the AI visibility category on 23 October 2026. For each one we read the robots file of the main host and of each candidate documentation host, classifying the response before reading a rule, because several of these hosts answer a 200 with a challenge page rather than a rules file. We then fetched at most one documentation page per vendor.

What the documentation pages showed is that the useful content is in an agent index rather than in the page itself. Nine of the 11 publish an llms.txt at their documentation host, listing 817 pages between them. We took that index as the evidence, because it is one request per host and it lists the documentation rather than one page of it. Each listed page was parsed into its title, url and description, and each field was credited on the title and description of a listed page.

Two filters are applied and both are stated on the page. Pages under an interface reference path are excluded, so an API endpoint is never counted as a statement about setting the product up. A duration only counts as a time to first report when the words around it are about a first result, so a description that says to send a report immediately is not read as a promise about setup time. Every figure on this page is a count taken from the recorded per-vendor records, and the indexes and the rules files are stored with their hashes so each count can be re-derived.