Optimising for ChatGPT is not a trick: it is making sure a machine can read you, understand you and trust you, in that order. If the first step fails, the other two do not exist.
1. Access: check you are not blocking yourself
Review your robots.txt. Many sites accidentally block assistant crawlers by copying old templates. Decide consciously whether to allow crawling: blocking protects content but removes you from answers. Also check that your main content does not depend on JavaScript to exist: if the raw HTML is empty, many crawlers see nothing.
2. Entity: make it obvious who you are
- One brand name spelled identically across site, social and directories.
- A clear category sentence on the home: what you do, for whom, where.
- Organization and product structured data linking your official profiles.
- An About page with verifiable facts: founding year, team, location, typical clients.
3. Format: write answers, not introductions
Models extract snippets. The winning snippet is short, self-contained and verifiable.
- Answer the question in the first two sentences of each section.
- One H2 per real question, phrased the way a customer would ask it.
- Lists and tables for comparable data: prices, lead times, limits, compatibility.
- Concrete numbers instead of adjectives: "replies within 4 hours" beats "fast support".
4. Trust: give it reasons to cite you
- Visible publish and update dates.
- Named authorship with real credentials.
- Linked sources whenever you claim something checkable.
- Consistency: if your price differs across three pages, AI will pick a more predictable brand.
5. Coverage: cover the full journey
Most sites only cover the sales pitch. Add the pages assistants actually use: honest comparisons, alternatives, explained pricing, use cases by sector, limitations and real FAQs.
6. Off-site: where the match is played
ChatGPT cites a lot of third-party content. Being in two or three reference industry roundups usually moves the needle more than any on-site tweak. Identify which sources your category is cited from today and work to appear in them.
7. Measure or you know nothing
None of these actions can be evaluated without a baseline. Define a fixed set of customer questions, measure weekly whether you are mentioned, in what position and which sources are cited, and compare before and after each change. Without a time series, optimising for ChatGPT is decorating blindfolded.