1. Establish exposure
Start at Google AI Overviews. It needs no input — the data comes from your existing rank checks. Read AIO Triggered against Total Keywords. That’s your trigger rate: how much of your keyword set now returns an AI answer. If it’s near zero, AI search isn’t affecting your market yet. Note the number, check again in a few months, and spend your effort elsewhere. Everything below assumes a meaningful trigger rate.2. Find where you’re losing
Two rates, and only one is yours to move. Trigger rate is out of your control. It’s exposure. Citation rate — of the queries that trigger an answer, how many cite you — is the target. Then sort the table by the situation that costs most: rows where you rank well but aren’t cited. Ranking third while an AI Overview above you cites three other sites means the answer is being given without you, and your position is delivering less traffic than it looks. These are also the most winnable, because you already have the authority.3. Find out who’s being cited instead
Expand those rows and read Referenced Sources. This is the step people skip, and it’s where the actual finding is. The cited sources are frequently not your competitors — they’re industry publications, documentation, community threads, and comparison sites. If those own your topic, editing your own page won’t change anything. Confirm the pattern at scale with Citation gap: up to 10 competitors against up to 50 keywords, with GAP flagged where they’re cited and you aren’t. Use Load saved competitors to pull your project list. Then check whether it’s universal. LLM benchmarking runs one query across several platforms at once:- Cited nowhere — a content problem. Compare the cited pages against yours.
- Cited on some platforms only — platforms weight sources differently. Look at what the platforms citing you have in common.
4. Fix the two different problems
If your own page isn’t good enough to quote, the patterns among cited pages are consistent:- The question is answered directly and early, not after several paragraphs of context
- The phrasing matches how people ask, not internal terminology
- Facts are stated plainly enough to be quoted in isolation
- Headings, lists, and tables make the relevant passage easy to locate
- A comparison site cited for your category that doesn’t list you — get listed
- A publication cited repeatedly — that’s a coverage target
- Community threads and forums — participate honestly; they’re heavily cited and rarely contested by vendors
5. Track it over time
One-off lookups diagnose; GEO measures progress. Add your key subjects as topics and they’re snapshotted daily by default. Phrase topics as questions where a vendor could plausibly be the answer. “best rank tracking tool for agencies” works; “rank tracking” returns a definition and will only ever produce zeros. On a topic’s detail page, watch two things:- Mentions and Citation Rate trending — mentioned without cited means you have awareness but aren’t treated as the source
- Model — citation behaviour shifts when a platform changes model. Check this column before attributing a step change to your own work
6. Expect variance
AI responses aren’t deterministic. The same query returns different sources on different runs. A single absence is not evidence. Re-run anything commercially important a few times over a week — consistent absence is a finding, a one-off is noise. This is the most common way to waste effort in this section.What to prioritize
If you only do three things:- Read the trigger rate on AI Overviews. It tells you whether any of this matters yet.
- Run Citation gap on your top commercial queries — where a decision is being made without you.
- Act on the sources, not only your own pages. That’s where most of the leverage is.