Skip to main content
AI-search optimization is the set of structural choices that let AI assistants find, cite, and reuse this handbook’s Attio Workflows guidance. An AI assistant reads a documentation page the way a hurried engineer does. It scans the title, the first sentence, and the nearest table. Then it either trusts the answer or moves on. This handbook is written for humans first. But every page is also shaped so a model can lift an accurate, attributed answer out of it. The goal is that it never invents a credit cost or makes up a block name. The payoff is practical for a go-to-market (GTM) engineer. Say a teammate asks an in-house assistant “does the Create record block cost a credit?” The model should return the real number and a link to the source page, not a confident guess. Structure is what makes that possible. A page earns a clean answer by doing three things. First, it answers the exact question in its first screen. Second, it names blocks with the exact labels Attio shows in the product. Third, it carries an official source link. Together these give the model a clear sentence to quote and a citation to attach. This page explains the conventions the handbook follows and how to apply them when you contribute. After reading it, you can do three things. You can write a page that an assistant can summarize correctly. You can audit an existing page for the patterns that stop an AI from finding the answer. And you can keep the machine-readable files in sync as Attio ships changes. (Machine-readable means written for software to parse, not just for people to read.)

​
Answering the question first

The single most useful move is to answer the exact question in the first screen, before any setup or background. Models lean hard on the opening sentence. The handbook makes that sentence carry the answer by reusing it as the page description. The description does three jobs at once. It is the meta tag, the search snippet, and the first line of body prose. So it has to read as a complete answer on its own. The Create record reference page shows the pattern in one line.
A model that retrieves only this frontmatter still answers “what does Create record do?” correctly. The answer never depended on the reader scrolling.

​
Naming blocks with stable labels

AI assistants match on names. So the handbook uses the exact label Attio shows on screen, never a paraphrase. Write Attribute value changed, not “the changed-field trigger.” Write Send HTTP request, not “the HTTP block.” Using the same name everywhere also helps a model connect two pages. When both pages call the trigger Record created, an assistant can link the build guide to the reference without guessing they mean the same thing. The same discipline applies to acronyms. Expand each one on first use, like GTM (go-to-market), ICP (ideal customer profile), and MCP (Model Context Protocol). That way a model that ingests a single page mid-document still has the definition in context. Compare the two phrasings a contributor might write for the same trigger. A page written with the right-hand column lets an assistant map every reference back to a real Attio block. That is the difference between a citable answer and a plausible-sounding one.

​
Front-loading cost and failure behavior

Cost and failure behavior belong near the top of a page. Those are the facts a model is most likely to be asked for, and most likely to get wrong. The handbook keeps one source of truth for these facts. The full credit table lives only on Credits & pricing explained. Every other page links there instead of repeating it. Stating one block’s cost inline is fine. So a contributor writes “1 credit” on a reference page, but never copies the whole table. Two costs need their exact wording so a model never rounds them off. The Execute code block costs Variable (complexity-based), not “free” and not “1 credit.” Logic blocks such as If and Filter consume no credits at all. A short inline cost note reads like this.
The number sits beside the block name. The link points at the one page that owns the table. So an assistant can quote the cost and still send the reader to the official source for plan-level math.

​
Including “when not to use this”

A page that shows only the smooth path teaches a model to recommend a block in the wrong cases. The model might suggest one that burns credits or overwrites good data. So the handbook adds a “when not to use this” note wherever a block is risky. The assistant then picks up the warning along with the feature. Calling out the limit is also the honest thing to do for a human reader skimming for a decision. The clearest example is the line between workflows and the API. A backfill is a one-time pass that fills in a field across many existing records. A bulk backfill across tens of thousands of records belongs in a script that calls the Attio REST API, not in a workflow. It belongs in a script because Find records returns at most 100 results.
A model that reads this note will steer a user toward the API for a migration. That is exactly the guidance a workflow page should carry. Without it, the page would quietly encourage a 50,000-record loop.

​
Linking concepts, references, and recipes

Internal links are how an assistant builds an answer that spans several pages. So the handbook connects every concept, reference, and recipe with root-relative links that have no trailing slash. The reference page for Round robin links to the recipe that uses it. The recipe links back to the block and out to the related concept. By following those links, a model can answer “how do I route inbound leads evenly?” It joins the block reference to the Round-robin lead routing build. The handbook also ships a machine-readable file alongside the prose. A /llms.txt file gives assistants one curated index of every route and what it is for.
An assistant that fetches /llms.txt gets the whole site map in one request. That lets it pick the right page to cite instead of guessing a URL.

​
Keeping pages source-backed

Every page that describes how the product behaves carries an official Attio source link. A sourced claim is one an assistant can attribute and a reader can check. AI-optimized does not mean confident-sounding. It means structured, source-backed, and easy to update when Attio ships a change. The handbook tracks the new June 2026 Workflows engine. In that engine, per-workflow credit caps were removed. You can no longer create new legacy workflows. A page that cites its source ages well as those details change. The honest source line at the foot of this page models the convention every page follows.
A model that quotes a behavior from a page can carry this line through as the citation. So the answer arrives with its source attached, instead of sounding like the assistant’s own claim.

​
See also

Source: Attio Help Center, Overview of workflows. Last reviewed against the Attio Help Center, June 2026.