Skip to main content
Documentation matters for SEO, GEO, and agent discovery because it connects a practical question to a concrete product capability. Existing users can use the answer to complete a task. Search engines can rank it for the task, and AI agents can retrieve it as evidence while evaluating or recommending the product. That means documentation creates value both before and after someone chooses the product. Before selection, it makes the capability findable and defensible. After selection, it helps a person or agent use the capability correctly without guessing. This lesson uses three observed examples to show how that happens:
  • Promptwatch found that how-to content and documentation together made up 8.2% of the ChatGPT citations it classified in its study period.
  • ChatGPT recommended Attio first for an agent-controlled CRM use case after searching Attio’s official documentation and citing its MCP capability.
  • A narrow Bun guide appeared fourth in a captured Google search for create simple HTTP server.
These examples do not prove that publishing documentation automatically produces rankings or recommendations. They show why documentation belongs in both the product-support strategy and the product-discovery strategy.

​
Documentation supports use, discovery, and product selection

By the end of this lesson, you should be able to:
  • explain the two moments in which documentation creates value;
  • describe why a focused documentation page is useful to a search engine or agent;
  • separate an observation from the conclusion it can reasonably support;
  • connect better documentation to a support or growth outcome;
  • recognize the pattern shared by the Promptwatch, Attio, and Bun examples.

​
How documentation creates value before and after product selection

Most teams recognize the value on the right side of this sequence:
Discoverable documentation can enter one step earlier:
This gives documentation two audience jobs and two organizational outcomes: The second outcome is possible because search engines and answer engines need sources. A product capability that exists only in a sales call, a dashboard label, or an internal explanation is difficult for an external system to retrieve and verify. A public page gives the capability a stable URL, a name, an explanation, and evidence.

​
Promptwatch: how-to and documentation made up 8.2% of classified ChatGPT citations

Promptwatch analyzed 1,027,802 classified ChatGPT citations from 1 January to 4 February 2026. In that sample:
  • How-To pages accounted for 5.9% of classified citations.
  • Documentation pages accounted for 2.3%.
  • Together, those two categories accounted for 8.2%, or roughly one in twelve classified citations.
Promptwatch chart titled What Type of Content Gets Cited by ChatGPT, showing How-To at 5.9 percent and Documentation at 2.3 percent from 1,027,802 classified citations.

Promptwatch's classified ChatGPT citations included 5.9% How-To content and 2.3% Documentation between 1 January and 4 February 2026.

​
What the Promptwatch citation data supports

How-to content and documentation form a meaningful part of the material ChatGPT cites. They are not only destinations for people who already know the product name. This makes sense from first principles. Many questions sent to an answer engine are practical:
  • How do I complete this task?
  • Which product supports this workflow?
  • What is the correct request or configuration?
  • Why did this operation fail?
A good how-to or documentation page is built to answer exactly that kind of question. It names the task, explains the mechanism, and provides steps or technical facts that an answer can use.

​
What the Promptwatch citation data does not prove

The chart does not show that documentation is the largest citation category, that every documentation page will be cited, or that changing a title will cause an 8.2% gain. Promptwatch also notes that its chart includes only citations it classified and that the mix of source types changes over time. Treat the result as evidence that these page types participate in AI discovery, not as a guaranteed ranking formula.

​
Attio: dedicated MCP documentation supported an agent recommendation

The Attio example begins with a broad product-selection question, not a branded search:
what crm can i use that an ai agent can control and help me do my crm tasks
In the captured result, ChatGPT considered Attio, HubSpot, Pipedrive, and Salesforce. It recommended Attio first for an AI-agent-centric setup and supported that recommendation with Attio’s official MCP integration, REST API, and webhooks. The result cited Attio’s documentation.
ChatGPT response to a question about a CRM an AI agent can control. The answer recommends Attio first, explains its official MCP integration, and shows an Attio Docs citation.

For a generic agent-controlled CRM question, the captured ChatGPT result recommended Attio first and cited Attio Docs as evidence for its MCP capability.

The relevant source is not a vague AI landing page. It is Attio’s dedicated MCP documentation page. The page makes the capability easy to understand and retrieve by answering a predictable sequence of questions:
  1. What is Attio MCP?
  2. Why would someone use it?
  3. What can an agent do with it?
  4. How is it connected?
  5. Which tools are supported?
  6. What could a user ask an agent to do?
  7. What are the rate limits and security conditions?
Its supported-tools table pairs machine-oriented tool names with plain-language descriptions and example prompts. An evaluator can move from “I need an agent-controlled CRM” to concrete operations such as searching, creating, updating, and deleting records.
Attio MCP documentation showing a Supported tools table with record and object tools, plain-language descriptions, and example prompts such as finding a person or creating a company.

Attio's MCP page maps supported tools to descriptions and example prompts, making the agent capability concrete rather than merely claiming AI support.

​
Why Attio’s MCP page is useful to an AI agent

An agent cannot reliably recommend a capability it cannot identify. The Attio page turns an abstract claim—“our CRM works with AI agents”—into a source that answers four retrieval questions: That specificity gives an answer engine something it can use to explain why the product fits the request. The documentation does more than make Attio mentionable; it makes the recommendation defensible.

​
What the Attio recommendation does not prove

This is one captured result. It does not prove that Attio will always be recommended ahead of HubSpot, or that the MCP page was the only cause of the recommendation. Product fit, brand knowledge, other sources, the model, and the wording of the prompt can all change the answer. The defensible lesson is narrower: when the model looked for evidence of an agent-controllable CRM, Attio had a dedicated first-party page that clearly described and demonstrated that capability.

​
Bun: a task-specific documentation page ranked for a generic search

The Bun example shows the same principle in traditional search. In the supplied Google result for create simple HTTP server, Bun’s guide ranked fourth with the title Write a simple HTTP server.
Google search results for create simple HTTP server, with Bun's Write a simple HTTP server documentation page shown as the fourth result.

In this captured Google result for create simple HTTP server, Bun's focused guide appeared in position four.

The result works without requiring the searcher to know Bun. Its title leads with the job the person wants to do. The product becomes the way to complete that job. Bun repeats this task-shaped pattern across focused guides: Compare the focused title Write a simple HTTP server with a broad title such as HTTP or Bun.serve reference. The focused version tells a searcher, search engine, or agent:
  • which task the page owns;
  • what outcome to expect;
  • why this page differs from neighboring HTTP pages;
  • which page to retrieve for this exact question.
The snippet reinforces the promise by describing the observable result: a server listening on a port and returning a response. The page is therefore useful in both stages. It can be selected for a generic task, and it can help the reader perform that task with Bun.

​
What the Bun search result does not prove

A search-result position is a snapshot. It can vary by date, location, device, and search-result changes. The screenshot does not isolate which ranking factor produced position four. What we can observe is that a product documentation page matched a non-branded task query with a narrow title and a page dedicated to the requested outcome.

​
How documentation enters search and agent discovery

The three examples operate at different levels, but they fit one discovery chain:
The page must succeed at both halves of the chain. A discoverable page that cannot deliver its promised task disappoints the reader. A complete page with a vague identity may never be selected. This is why the course evaluates documentation with two verbs:
  • Find: Can a person, search engine, or agent identify the right page for the question?
  • Use: After selecting it, can a person or agent complete or verify the answer?

​
Audit one product capability for documentation discovery

Choose one important product capability and answer these questions:
If no single page owns the answer, the capability is harder to discover and harder to use. The next lesson gives you the model for evaluating what that page must do. Continue to What Makes a Good Documentation Page.