How to get cited by AI search
AI answers cite sources. Being one of them is mostly unglamorous engineering plus a content decision most people get backwards.
First, be reachable
An AI assistant cites what its index contains. If your pages are not crawled, nothing else matters. The failures we see most often are boring:
- Pages that only exist after JavaScript runs
- A sitemap declared in robots.txt that does not exist
- Missing pages returning a rendered error page with an HTTP 200, so every bad URL looks like real content
- Blanket crawler blocks copied from a template, quietly excluding the exact agents you want
Check the boring things before optimising anything.
Be the best answer to a narrow question
This is where most effort goes wrong. Broad, popular topics are the most competitive and the least winnable for a small site. Narrow, specific, badly served questions are winnable immediately.
The test is simple. Imagine the exact sentence someone types into an assistant. If a dozen established sites already answer it well, you will not be cited. If the honest answer today is a forum post from three years ago and a GitHub issue with four replies, you can be the best source in an afternoon.
Answer, do not tease
Content written to make someone click, subscribe or enquire withholds the answer by design. That is precisely the content an assistant cannot use. Put the answer in the page, completely, including the parts that make you less necessary.
Structure helps too. A clear question as a heading with the answer directly underneath is easy to extract. A meandering narrative with the answer in paragraph nine is not.
Be specific enough to be verifiable
Versions, exact error text, real numbers, the actual command. Specificity is what separates a page that gets cited from one that gets skimmed, because it demonstrates the content came from doing rather than summarising.
Get new pages in fast
Waiting for a crawl wastes days. Submit new and changed URLs directly where the platform supports it, keep a real sitemap updated, and link new pages from somewhere already crawled regularly. See our note on IndexNow.
Measure what actually happened
There are two very different kinds of tool. Some track prompts you define and report how you score. Others report genuinely observed citations, where a real answer cited a real page of yours.
Only the second tells you anything about the world. If your reporting improves the day after you added prompts, you measured your own prompt list, not your visibility. Know which you are looking at before drawing conclusions.
Need help with any of this?
These notes are free and always will be. If you would rather someone just set it up, or you are stuck on something similar, get in touch at hello@opsira.io.