Do these in order, and stop when it works
Content optimisation for AI has three layers and they are not equally likely to be your problem. Working them out of order is how teams spend a quarter rewriting pages that an assistant was never able to fetch.
| Layer | The question | How often it is the problem |
|---|---|---|
| Delivery | Does the content arrive at all? | Most often, and least suspected |
| Structure | Can the facts be found once it arrives? | Frequently |
| Wording | Is it phrased the way people ask? | Real, but rarely the blocker |
Layer one: delivery
- Is the fetch allowed? Read every
robots.txtline and CDN bot rule and ask whether you would write it today. Most assistant blocks were inherited, not chosen. - Does it survive without JavaScript? Prices, hours, fees and eligibility that render client-side are absent from what an agent receives.
- Is the response affordable? Agents read under a token budget. A page that is mostly theme spends its budget before reaching your content.
The fix for all three is serving-layer work rather than content work, which is why it belongs to engineering and takes an afternoon rather than a quarter.
Layers two and three
Once the content arrives, structure decides whether the useful part survives summarisation. What helps a parser is unglamorous: the answer stated near the top, facts as facts rather than implied by design, and structured data that agrees with the prose. What does not help is heading density, keyword placement or word count, because there is no position in a generated answer to compete for.
- Put the answer first. For each template, find the offset of the first fact a person would ask about, and cut what sits above it.
- State facts explicitly. A price in an image, a fee in a tooltip and hours in a widget are all invisible.
- Make the markup agree. Fix disagreements between prose and structured data rather than adding more markup.
- Then write for the question. Phrase headings as the questions people actually ask, and answer them in the first sentence beneath.