How to Optimize for ChatGPT: a Page-Level Checklist

Optimizing for ChatGPT means making every key page answer its question in the first two or three sentences, keeping your brand facts identical across site, schema and llms.txt, allowing OpenAI’s crawlers, and earning mentions in the sources ChatGPT cites. Below is the page-level checklist we apply to our own site and to every engagement.

Our guide on how to rank in ChatGPT covers the strategy. This post is the execution layer: what to change on the actual pages, in the order we do it.

The crawler layer: five minutes, do it first

Check robots.txt allows GPTBot (training crawler), OAI-SearchBot (the search index behind ChatGPT’s browsing) and ChatGPT-User (live user requests). Blocking OAI-SearchBot removes you from search-backed answers entirely, and plenty of sites block it by accident through a blanket rule.

Then verify your key pages render as HTML without JavaScript. Fetch them with curl or any text-only viewer: if the pricing and the product description are not in the raw HTML, assume the engine sees a blank page.

The entity block: say who you are like a database would

Every optimized site has one canonical sentence: “[Brand] is a [category] for [audience] that [outcome].” Ours: Sourcelift is a productized GEO agency for B2B brands that gets them cited inside AI answers. That sentence, unchanged, belongs in the homepage opening, the about page, the meta descriptions, the Organization schema and llms.txt.

Engines cross-check. Every variation between your homepage and your LinkedIn is a small doubt, and doubt loses citations to whoever is unambiguous.

The answer-first rewrite, page by page

For each page mapped to a buying question, apply the same structure: H1 that contains the question’s language; an opening block of two or three self-contained sentences that answer it completely (a reader who stops there should have the answer); then sections that each cover exactly one subquestion with a descriptive H2; then an FAQ with the questions buyers actually append.

Two details that matter more than they look. Keep one H1 per page, because engines use heading hierarchy to segment meaning. And write FAQ answers as complete standalone paragraphs, because they get lifted individually.

Schema: the four types that pull weight

Organization (or ProfessionalService) once, sitewide, with the same facts as your entity block. Service or Product on offer pages, with real prices if you publish them. FAQPage wherever a real FAQ exists. Article on blog posts, with dates. Validate with Google’s Rich Results test; broken schema is worse than none because it signals carelessness about exactly the thing you want trusted.

llms.txt: your brand, pre-summarized

llms.txt is a markdown file at your root that gives language models a curated summary: who you are, key facts, and links to the pages that matter with one-line descriptions. Adoption by engines is uneven and honest practitioners say so. We publish one anyway (here is ours) because it costs an hour, it is the cheapest insurance available, and it forces the entity clarity work that pays off everywhere else.

Freshness and consistency, the boring multipliers

ChatGPT’s search prefers pages that look maintained: visible dates, updated pricing, current screenshots. A comparison page from 2024 loses to a thinner one from last month more often than seems fair. Schedule a quarterly pass over your money pages, and update the dateModified in schema when you do it for real. Never fake the date; engines compare content diffs.

Prove it moved

The checklist above is falsifiable, which is the point. Fix a prompt set of 20 to 40 buying questions, run it before you touch anything, run it again monthly, and score recommended, mentioned or absent per prompt. That is the same instrument as our Visibility Baseline, scaled to a spreadsheet.

If you want the baseline done for you, start with the free AI Visibility Diagnosis. If you want the whole checklist executed with the measurement attached, that is the 90-day sprint.