Search Console gives you four numbers per row: clicks, impressions, position, CTR — plus the query and page dimensions behind them. Ten years of SEO content taught you how to improve those numbers. This guide teaches you how to read them for a different game: Generative Engine Optimization — deciding which pages and questions AI answer engines should be citing you for.
The reframe in one line: Search Console stops being a scoreboard and becomes an opportunity map.
Impressions: the citation opportunity map
The metric most teams underrate is the one GEO needs most. An impression means your page appeared for a query — including queries where an AI Overview or AI Mode answered the user and they never clicked anything. Those answer-satisfied queries are exactly where citations are being distributed, and impressions are the only metric that still counts them.
The reading: Performance → Search results → Pages → sort by impressions, 28-day window. Your top 20 pages are your GEO scope. Why 20: it's the slice where authority demonstrably exists (impressions prove retrieval and trust), and it's a workable batch for content rewrites. Below ~20 pages, effort disperses; above it, you're usually looking at pages too weak to be cited this cycle.
The second reading — the impression-to-click gap. Sort by impressions, then scan for high-impression / near-zero-CTR rows. Each row is a question your audience asked that got answered without you — the user read an AI answer (someone else's citation) and left. Every such row is either your next citation win or a competitor's current trophy. The AI-visibility audit tells you which.
Position 5–20: the golden zone, precisely defined
Position data earns its GEO value in a specific band. Pages ranking 5–20 satisfy three conditions at once:
- Retrieved — they're in the candidate pool answer engines draw from
- Trusted — Google's ranking system has validated topical relevance and authority
- Structurally unfinished — they rank despite not being extraction-ready, which means a structural rewrite (not a link campaign) is the missing piece
Below position 20, authority work (links, depth) usually must come first — GEO can't cite what retrieval won't surface. Above position 5, you're already winning the SERP click; the GEO question is only whether the AI answer cites you too.
The reading: Queries tab → filter position > 5 and < 20 → sort by impressions. Every row is a page that's one rewrite from being cited. This is the position 5–20 method, and it's the default filter in the Scrabio analysis precisely because the band math works.
CTR: the X-ray of where the click died
CTR is impressions' truth serum. Read it in both directions:
Low CTR + high impressions → the query was satisfied without a click. In 2026 that usually means an AI answer did the satisfying. The SEO instinct ("improve the title tag!") is often wrong here — you'd be optimizing a snippet the user never saw because the AI answer absorbed the attention. The GEO instinct is correct: check who got cited, and become the source.
High CTR → you're winning the click AND presumably competing for the citation. These pages need defense, not rewrites — keep passages fresh, because citation ownership decays quietly.
The trap to avoid: aggregate CTR targets. "Raise site CTR to 5%" is meaningless when the honest decomposition is "raise CTR where the SERP still shows links, and replace lost clicks with citations where it doesn't."
Queries: your audience's verbatim language — the sleeper asset
Connect GSC to your AI
Query your Search Console data conversationally in Claude or ChatGPT — free MCP server.
Get the free GSC MCPThe query dimension records what humans actually typed: fragments, full questions, dialect, slang, year modifiers, mixed languages. Nothing else in your stack — keyword tools included — records the verbatim layer at this fidelity, tied to your pages.
Four extractions that feed GEO directly:
- Question shapes — full interrogatives ("what is the best hair oil for a dry scalp?") vs fragments ("best hair oil dry scalp"). The dominant shape per page decides your heading-vs-FAQ format.
- Register and dialect — Egyptian Arabic vs MSA, casual vs clinical. AI engines match the asker's register; content in the brand's default register loses the citation to content in the asker's.
- Intent splits — "best X" (comparison), "why does X" (troubleshooting), "how to X" (instructional) each become their own passage with its own heading.
- Modifiers as structure — "for curly hair", "without heat", "2026" are section titles waiting to exist.
The full mining method is in our query-language guide. The one-line version: your FAQ questions should be your users' queries, verbatim, with grammar cleaned but phrasing preserved.
The 15-minute property reading
Here's the complete GEO reading of any Search Console property, on a timer:
| Minutes | Action | Output |
|---|---|---|
| 0–3 | Pages → sort by impressions → top 20 | The GEO scope list |
| 3–6 | Queries → filter position 5–20 → sort by impressions | The rewrite target list |
| 6–9 | For each of the top 5 pages: pull its queries verbatim | The question set / content language guide |
| 9–12 | High impressions + near-zero CTR rows | The "answer-satisfied" list → citation check needed |
| 12–15 | High CTR pages | The defend list → keep passages fresh |
Fifteen minutes produces four working lists: scope, targets, gaps, defenders. That's a month of prioritized GEO work.
Do it conversationally instead
Every step in that table is a question you can ask your AI directly. Connect your Search Console to Claude or ChatGPT with our free MCP server, then:
- "Top 20 pages by impressions, last 28 days, with average position"
- "Queries ranking in positions 5–20, sorted by impressions"
- "All queries for /best-hair-oil, verbatim"
- "Pages with more than 1,000 impressions and CTR under 1%"
And when the reading is done and you want the rewrites executed — one GEO analysis run turns the four lists into published-ready, citation-scored content. The data was always there. Now it has a second job.
