The GEO Audit: What Clients Are Asking Agencies For (Template Included)

A GEO audit is a structured review of whether AI search engines cite a brand when someone asks about its category, and why they do or don't. It samples the questions real buyers type into ChatGPT, Perplexity, Gemini, and Google's AI answers, records who gets named in the responses, and maps the specific reasons a brand is present or absent so a team can fix the gaps in priority order.

The reason you need one on file: your clients have started asking a question you can't answer from a rankings dashboard. "What do we look like in ChatGPT?" "A prospect said the AI recommended a competitor. Why not us?" Traditional SEO reporting doesn't cover it, because AI answers don't behave like the ten blue links. A page can rank third on Google and never get cited by an AI engine. Another can sit on page two and get quoted verbatim.

This is the audit that closes that gap. Below is the full section-by-section structure, written so you can copy it into your own client deliverable. Each section says what to check, how to check it by hand, what a good result looks like, and where Frase does the same work faster. If you want the deeper background on the category first, start with what generative engine optimization is.

What is a GEO audit, and how is it different from an SEO audit?

An SEO audit asks: can this site rank, and does it? It checks crawlability, on-page signals, backlinks, and keyword positions against Google's index. A GEO audit asks a different question: when an AI engine assembles an answer, does it pull this brand in as a source, and does it describe the brand correctly?

The mechanics differ because the surface differs. Google returns a ranked list. AI engines return a synthesized answer built from a handful of sources they judge trustworthy and easy to extract. A brand can win on the first and lose on the second. That is why a GEO audit samples answers instead of rankings, records citations instead of positions, and weighs extraction-readiness and authority signals that a standard crawl ignores.

Run both. They share a foundation, and a lot of what earns AI citations also helps traditional rankings. The GEO audit sits on top and answers the client's newer question.

Section 1: Baseline citation check across AI engines

What to check. Whether the brand appears at all in AI-generated answers for its most important category questions, and across which engines. This is your before-picture. Everything later gets measured against it.

How to check it manually. Build a list of 15 to 25 questions a buyer would actually ask, then run each one through ChatGPT, Perplexity, Gemini, and Google's AI Overviews. For every answer, log three things: was the brand cited, which sources were cited instead, and how the brand was described if it appeared. A simple spreadsheet with one row per question and one column per engine works. Note that AI answers vary between runs, so check each question two or three times and record how often the brand shows up, not just whether it did once.

What "good" looks like. The brand is cited in a meaningful share of category answers on at least the engines its buyers use, and the description is accurate. Total absence across every engine is the most common starting point, and it's a fixable one.

How Frase does it faster. Running two dozen questions across several engines by hand, several times each, is a half-day of tab-switching. Frase AI Visibility runs the tracking across AI engines on a schedule, records where the brand is cited, and flags changes over time. You keep the same baseline picture without rebuilding the spreadsheet every month.

Section 2: Branded vs category query sampling

What to check. The gap between how AI engines answer questions *about the brand by name and questions about the category* where the brand should surface unprompted. These are two different battles.

How to check it manually. Split your question list. Branded queries name the client directly ("Is [Brand] good for X?", "What does [Brand] do?"). Category queries describe the need without naming anyone ("best tool for X", "how do teams handle Y"). Run both sets. Branded answers tell you whether the AI has accurate information about the client. Category answers tell you whether the client is in the consideration set at all when a buyer hasn't heard of them yet.

What "good" looks like. Branded answers are accurate and on-message. Category answers name the client alongside the obvious incumbents. Most clients are stronger on branded than category, which points the work toward category-level authority.

How Frase does it faster. Frase tracks both branded and category prompts so you see the two trends side by side rather than assembling them from separate manual passes.

Section 3: Content extraction-readiness review

What to check. Whether the brand's own pages are structured so an AI engine can lift a clean, quotable answer from them. AI engines favor content they can extract without ambiguity.

How to check it manually. Take the pages that should be earning citations and read them the way a machine would. Does each page answer its core question in the first paragraph, in plain language, before the marketing wind-up? Are there clear headings phrased as the questions people ask? Are claims stated as self-contained sentences that make sense pulled out of context? Long, buried, or hedged answers are hard to extract. A page that opens with a direct one-sentence answer and supports it underneath is easy to extract. This is the single biggest on-page lever, and it's covered in depth in the complete GEO playbook on getting cited by AI.

What "good" looks like. The most important pages lead with the answer, use question-shaped headings, and state facts in liftable sentences. A reader (or a model) gets the answer in the first few lines.

How Frase does it faster. Frase scores content against the signals AI engines look for and shows what to change on a given page. Instead of eyeballing extraction-readiness page by page, you get a score and specific edits, then rewrite against them.

Section 4: Schema and technical pass

What to check. The structured-data and technical signals that help engines parse a page and trust it, including markup, crawlability, and clean HTML.

How to check it manually. Check whether key pages carry appropriate schema (FAQPage, Article, Organization, Product where relevant), whether that markup validates, and whether the content an engine needs is present in the served HTML rather than locked behind scripts. FAQ schema in particular maps neatly onto the question-and-answer shape of AI results. There's a focused walkthrough in how FAQ schema supports AI search. Also confirm the basics: pages are crawlable, canonical tags are sane, and important answers aren't hidden from parsers.

What "good" looks like. Priority pages carry valid, relevant schema, serve their core content in HTML, and are cleanly crawlable. Schema won't manufacture authority on its own, but its absence makes a page harder to parse and quote.

How Frase does it faster. Frase surfaces technical and structure issues as part of its audit so the schema and extraction gaps land in one place instead of across separate validator tools.

Section 5: Off-site authority signals

What to check. The third-party signals that make an AI engine trust a brand enough to cite it: mentions on authoritative sites, presence in the roundups and listicles that engines lean on, reviews, and consistent descriptions of the brand across the web.

How to check it manually. Go back to your baseline citation log and note which *sources* the engines cited instead of your client. Those recurring domains, often review sites, comparison articles, and category roundups, are the authorities the engines trust for this topic. Then check whether the client appears on them at all. Also check that the brand is described consistently wherever it shows up. AI engines synthesize from many sources; contradictory descriptions weaken the picture.

What "good" looks like. The client is present and accurately described on the third-party sources the engines actually cite for its category. The most common finding here is that a few authority domains dominate the citations and the client is missing from most of them. That gap is your off-site roadmap.

How Frase does it faster. Frase's tracking shows which sources get cited for the prompts you care about, which points you straight at the authority domains worth earning placement on. The outreach itself stays manual, but the target list stops being guesswork.

Section 6: Competitor citation comparison

What to check. Who *is* getting cited when your client isn't, and why. This is the section clients react to most, because it turns an abstract worry into a named scoreboard.

How to check it manually. From your baseline log, tally citations by brand across your category questions. Build a simple table: competitor name, how many of your category answers cited them, and a note on what those citations had in common (a strong comparison page, a review-site presence, a frequently-quoted stat). The pattern usually inverts the client's own sense of the market. A brand that feels like a leader by revenue or traffic can be nearly invisible in AI answers, while a smaller competitor with extraction-ready content and review-site presence gets quoted constantly.

What "good" looks like. Your client holds citation share comparable to its real market position, and you can explain any competitor's lead by a specific, copyable reason rather than mystery.

How Frase does it faster. Frase tracks competitor presence in AI answers alongside your client's, so the scoreboard updates on its own instead of being rebuilt by hand each reporting cycle.

Section 7: Prioritized fix roadmap

What to check. Not a check. This is the synthesis. Every gap above becomes a ranked action with an owner and an effort estimate.

How to build it. Sort findings by impact over effort. Extraction-readiness rewrites on pages that already have traffic and topical relevance tend to be the fastest wins, because the authority is already there and only the structure is holding the page back. Off-site authority placements are higher effort and slower to pay off but compound. Schema fixes are usually low effort. For each item, write the gap, the recommended fix, the effort, and the expected result, then group them into "recoverable now," "build over the quarter," and "long-term authority." Structure the whole deliverable as Strengths, then Gaps, then Path. Lead with where the client already wins so the gaps read as opportunity, not indictment.

How Frase helps. Frase's content scoring turns "improve this page" into specific edits, so the top of your roadmap moves from diagnosis to done inside the same tool. Content Guard is a separate feature that watches Google ranking and traffic decay over time and proposes fixes for pages that slip, applying them once a human approves (or automatically only if you opt in). It's a maintenance layer, not part of the AI-citation audit, but worth knowing about when a client asks how you'll keep gains from eroding.

The template: copy this and run it

Here is the whole audit as a fill-in checklist. Copy it into your own doc, one per client.

GEO AUDIT — [Client] — [Date]
Auditor: ____________  Engines checked: ChatGPT / Perplexity / Gemini / AI Overviews

1. BASELINE CITATION CHECK
   [ ] 10 buyer-intent queries drafted (no brand names in query)
   [ ] Each query run on every engine; cited yes/no logged per engine
   [ ] Result: cited in ___ of ___ query-engine pairs

2. BRANDED VS CATEGORY SAMPLING
   [ ] 5 branded queries run (is the brand described accurately?)
   [ ] 5 category queries run (does the brand appear at all?)
   [ ] Misdescriptions found: _______________________________

3. CONTENT EXTRACTION-READINESS
   [ ] Top 5 revenue pages reviewed for a direct answer in the first two sentences
   [ ] Question-shaped headings present? Per page: ____________
   [ ] Pages needing a rewrite-for-extraction: ________________

4. SCHEMA / TECHNICAL PASS
   [ ] Article/FAQ schema present and valid on key pages
   [ ] Pages render without JS for crawlers (view-source test)
   [ ] Canonical + robots clean on the pages that matter

5. OFF-SITE AUTHORITY
   [ ] Brand present in the third-party articles engines cite for the category?
   [ ] Review-site profiles current (G2/Capterra/Trustpilot)?
   [ ] Gaps to pitch: _______________________________________

6. COMPETITOR CITATION COMPARISON
   [ ] Same 10 baseline queries: which competitors ARE cited?
   [ ] Pattern: what do their cited pages do that ours don't?

7. PRIORITIZED FIX ROADMAP (Strengths → Gaps → Path)
   [ ] Top 3 strengths worth defending: _____________________
   [ ] Top 5 gaps ranked by revenue relevance: ______________
   [ ] 30/60/90-day plan drafted with owner per fix

REPORTING: re-run monthly · re-run 2 weeks after each fix batch
KPIs: citation rate on the 10-query set · engines covered · misdescription count

That is the deliverable clients are asking for. The sections above explain how to fill in each line well.

How often should you re-run a GEO audit, and what do you report?

Cadence. Run the full audit at engagement kickoff to set the baseline, then re-sample the citation check monthly and run the full pass quarterly. AI answers shift as engines update and as competitors publish, so a static one-time audit goes stale. Monthly re-sampling catches movement; the quarterly full pass catches structural change.

What to report. Keep the client-facing report to a handful of KPIs they'll remember: citation presence (share of category questions where the brand is cited, by engine), branded-answer accuracy, competitor citation share, extraction-readiness score on priority pages, and roadmap progress. Report presence as a trend line, not a single number, so the client sees direction. Scale plans support multi-site tracking, multiple seats, and exportable reports, so agencies can pull a clean, client-ready report per account rather than hand-assembling one. Full white-label client portals with custom domains sit on Enterprise.

Set expectations up front. A GEO audit finds the gaps and the client can start closing extraction and schema issues within a cycle. Off-site authority and category-level citation share build over quarters, not weeks.

The fastest way to run your first one

You can run this entire template by hand. Many agencies do, and the manual pass is worth doing once because it teaches you how the engines actually behave for your client's category. The cost is time: a full manual audit across engines, with repeat sampling, runs the better part of a day per client, and it goes stale the moment you finish.

The alternative is to run the baseline citation check, competitor comparison, and content scoring inside a Frase trial and keep the manual review for the judgment calls. Point Frase at a client domain, pull the AI Visibility baseline and GEO scores, and you have the first three sections of the deliverable drafted before lunch. It's also the strongest way to show a prospective client what an ongoing engagement looks like: run their audit live, on their own domain, in the first call.

For agencies building this into a repeatable service line, the agency solutions overview covers how the multi-site and reporting pieces fit together.

---

FAQ

What is a GEO audit?

A GEO audit is a structured review of whether AI search engines cite a brand when people ask about its category. It samples real buyer questions across engines like ChatGPT, Perplexity, Gemini, and Google's AI answers, records who gets cited, and maps the specific reasons a brand is present or absent so a team can fix the gaps in priority order.

How is a GEO audit different from an SEO audit?

An SEO audit checks whether a site can rank in Google's list of results. A GEO audit checks whether AI engines pull the brand in as a cited source when they synthesize an answer. A page can rank well and still never get cited, so the two audits measure different things. Most teams run both.

What tools do I need to run a GEO audit?

At minimum, access to the AI engines your client's buyers use and a spreadsheet to log citations. That's enough for a manual baseline. To track presence over time, compare against competitors, and score pages for extraction-readiness without rebuilding the work each month, a tool like Frase handles the tracking and content scoring so you focus on strategy.

How long does a GEO audit take?

A thorough manual audit across engines, with repeat sampling for consistency, takes the better part of a day per client. Using a tool to handle the citation tracking and content scoring cuts the baseline to a few hours, leaving your time for the judgment calls and the roadmap.

How often should I re-run a GEO audit for a client?

Set the baseline at kickoff, re-sample the citation check monthly, and run the full audit quarterly. AI answers shift as engines update and competitors publish, so monthly checks catch movement and the quarterly pass catches structural change.

What should a GEO audit report include for a client?

Keep it to a few memorable KPIs: citation presence by engine shown as a trend, branded-answer accuracy, competitor citation share, extraction-readiness scores on priority pages, and roadmap progress. Structure the narrative as Strengths, then Gaps, then Path, so the client sees where they already win before they see the work ahead.

Can I show a client their AI visibility during a sales call?

Yes, and it's one of the most effective ways to open an engagement. Running a live baseline citation check on the prospect's own domain turns an abstract worry into a concrete, named picture of where they stand against competitors, which is far more persuasive than a generic pitch.