Relevance engineering
Relevance engineering is shaping your content so AI systems can find it, read it and judge it the best match for a question.
Relevance engineering is a name for building content around how AI systems decide what's relevant. You work on what every system does, whatever its algorithm: finding passages that match a question, judging which are best, and using them in an answer.
It's a useful way to think because it gives you causes you can work on. The platforms change their models all the time. How they search, read and pick passages changes much more slowly.
What it covers
The work usually falls into a few parts:
- Making sure AI bots can reach and read your pages, with the text in the HTML
- Writing sections that each answer one question and make sense on their own
- Using the words your customers use, so a passage matches what they ask
- Making clear which brand, product and place each page is about
- Earning mentions on the other sites the platforms read
Most of this overlaps with good SEO and good writing. The difference is the unit you think in. Search engines rank pages. AI systems often pick out passages, so each section has to stand on its own.
An example
Kestrel Roasters has one long page about brewing. A question about "how much coffee for a 1 litre cafetière" matches one paragraph halfway down, under a heading that just says "Tips." Kestrel splits the page into sections with headings that say what each answers, and puts the answer to each in its first sentence. Now that paragraph is easy to find and easy to quote.
In AnswerOrbit
Page checks in Recommendations runs eight checks on a page for a prompt, such as whether AI crawlers may read it, whether it's shown without scripts, and whether its opening answers the prompt. It runs the same checks on up to five pages the platforms cite for that prompt, so you can compare.