How to Run a GEO Audit in AI Search: Checklist, Scoring, Priorities
A GEO audit checklist for your own site: nine checks, a 100-point score and a priority order, with canlah.ai audited against it and scoring 71.
Quick answer
Canlah AI runs a nine-check GEO audit checklist in this guide: query universe, technical crawl and render, AI citation and funnel position, entity and structured data, third-party listings, off-site authority, conversion and attribution, on-page trust signals and community mentions. A GEO audit checklist is a fixed list of checks you run against your own website to find out why AI engines do or do not cite it. Each check produces a recorded number with its denominator, and the nine numbers roll into one score out of 100.
Canlah AI ran this checklist against canlah.ai on September 7, 2026 and scored its own site 71 out of 100, with 0 blocking issues, 8 high-priority issues and 14 medium-priority ones. The weakest check was AI citation and funnel position at 4 out of 100: across 177 grounded answers to open buyer questions, the site was mentioned 7 times, a 4.0% rate, while all 39 grounded answers to questions that named the brand mentioned it.
Treat a single audit as directional. One run cannot prove stable share of voice, causation or future recommendations. If the audit reveals a real gap, lock the question pool and repeat the same checks by engine, market and language before deciding what to change.
Disclosure: Canlah AI publishes this guide and sells the managed audit described below. The canlah.ai figures come from a first-party self-audit export dated September 7, 2026 that has not been independently verified. Tool capabilities and prices come from public product pages reviewed on September 23, 2026 and were not confirmed through paid accounts.
What does a GEO audit measure?
A GEO audit measures two things that buyers often merge into one number: whether AI engines currently name your brand, and whether your site is readable enough for them to do so later. The first is an answer-side measurement and changes every time you run it. The second is a site-side measurement and stays stable until you ship a change.
- Mention rate: the share of grounded answers to a fixed question pool that name the brand at all.
- Funnel position: whether the brand appears on category questions, on shortlist questions or only on questions that already name it.
- Citation: whether the answer links to your own domain, a third-party page, a competitor or no visible source.
- Crawlability: whether engines and their crawlers can reach, render and parse the pages that hold the answer.
- Entity clarity: whether Wikidata, structured data and your own facts page agree on what the company is.
- Source ecosystem: which domains the engines cite instead of yours, and whether you can be published on them.
The nine checks in a GEO audit checklist
A GEO audit checklist has nine checks, and canlah.ai scored between 4 and 90 on the eight that were scored on September 7, 2026, for a composite of 71 out of 100.
The nine-check GEO audit checklist, with canlah.ai scored against it on September 7, 2026 (composite 71 of 100).
| # | Check | What to record | canlah.ai score | Decision it supports |
|---|---|---|---|---|
| 1 | Query universe | Candidate buyer questions found, questions probed, questions pruned as irrelevant | 80 of 100 (191 found, 34 pruned, 39 probed) | Whether you are measuring questions buyers actually ask |
| 2 | Technical crawl and render | Pages crawled, sitemap URLs, orphan pages, render mode used | 85 of 100 (80 pages, 174 sitemap URLs, 17 orphan pages) | Whether engines can reach the pages that hold your answers |
| 3 | AI citation and funnel position | Mentions per layer, per engine, per run, with the cited domains | 4 of 100 (7 of 177 open answers: 7 of 84 OpenAI, 0 of 93 Gemini; 39 of 39 branded) | Where in the buying funnel you become invisible |
| 4 | Entity and structured data | Wikidata item, Wikipedia article, schema types present on key pages | 60 of 100 (0 Wikidata candidates, no Wikipedia article) | Whether engines can resolve who the company is |
| 5 | Third-party listings | Directory and review-platform records that verify the brand | Not scored (0 of 1 Clutch records found) | Whether an independent record exists to cite |
| 6 | Off-site authority | External mentions, unlinked mentions, source types | 65 of 100 (7 mentions, 3 unlinked) | Which earned coverage is worth converting first |
| 7 | Conversion and attribution | Enquiry forms found, analytics container detected, event names | 85 of 100 (no enquiry form, no container ID detected) | Whether AI-sourced demand can be measured at all |
| 8 | On-page trust signals | Specification tables, comparison blocks, named authorship, dates | 90 of 100 (specification and comparison blocks missing) | Whether a page answers a comparison question completely |
| 9 | Community and conversational mentions | Mentions per platform, subreddits reached, sentiment split | 90 of 100 (11 mentions, 8 on Reddit across 7 subreddits) | Whether real discussion exists for engines to retrieve |
Source: Canlah AI self-audit export for canlah.ai, September 7, 2026. Scores are the pipeline’s own dimension scores, first-party and not independently audited. “Not scored” means the check was run but the pipeline excluded it from the composite under its rule for B2B businesses.
The composite for canlah.ai was 71 out of 100, and a check scoring 90 does not cancel a check scoring 4. That is the main reason to keep all nine scores visible rather than reporting only the composite. The exclusion of check 5 is also worth challenging: a B2B buyer shortlisting agencies may still read Clutch, so an empty record there is a gap even when the pipeline does not score it.
How to run a GEO audit, step by step
A GEO audit runs in nine steps, answer side first: steps 1 to 5 cover checks 1 and 3, and steps 6 to 9 cover the site-side and off-site checks.
- Build the query universe before touching the site (check 1). List the questions buyers type in discovery, comparison and decision stages, and keep the branded questions in a separate layer.
- Prune the universe against the business. The Canlah AI run on canlah.ai discarded 34 of 191 candidate questions as irrelevant before a single probe ran, then probed 39 across six funnel layers.
- Probe each surviving question on every engine your buyers use, more than once (check 3). The canlah.ai run sent 39 questions three times each to two engines, for 234 planned answers.
- Record engine, market, language, account state, date and time for every run, then archive the full answer and the cited URLs. In the canlah.ai run 233 of 234 answers succeeded and 17 had no grounding sources, so the scored denominator was 216 grounded answers.
- Score mentions by funnel layer, not in aggregate. A 100% branded mention rate beside a 4.0% open mention rate is a positioning finding, not a measurement error.
- Crawl the site and compare the crawled page list against the sitemap (checks 2 and 8). Note orphan pages, render failures, missing comparison or specification blocks and any page that needs JavaScript to show its main answer.
- Check entity records (check 4): Wikidata item, Wikipedia article, structured data types and whether your own facts page repeats the same company description.
- Collect off-site evidence (checks 5, 6 and 9): external mentions, unlinked mentions, directory records and community threads, with the source domain for each.
- Check that demand can be measured (check 7). Confirm that an enquiry form exists, that the analytics container ID is present in the served HTML and that lead events fire.
Steps 1 to 5 are the part that breaks when hand-rolled, because engine access, sampling discipline and archiving all drift. Steps 6 to 9 can be run with free tools on a quarterly cadence.
How to score a GEO audit checklist
A GEO audit checklist is scored check by check out of 100, with non-applicable checks excluded, and the canlah.ai audit returned 0 blocking issues, 8 high-priority issues, 14 medium-priority issues and 2 passing checks behind its composite of 71. Those issue counts are a more actionable summary than the composite, so read them first.
Three scoring rules keep the result defensible. Excluded checks must be marked as excluded rather than scored zero, because a zero that means “not measured” and a zero that means “measured and absent” lead to different work. Every score needs the denominator beside it, so 4 of 100 reads as 7 mentions in 177 answers. Any check whose evidence was collected through an API rather than a signed-in consumer interface should be labelled: the canlah.ai run queried OpenAI through the API rather than the ChatGPT app, and its Gemini leg used a floating model alias, so the exact model version was not recorded.
How to prioritise what a GEO audit finds
A GEO audit’s findings are prioritised in three monthly stages: attribution fixes in month one, landing pages in month two and off-site authority in month three. The canlah.ai audit applied that order to its own findings, assigning month one to the two attribution gaps, month two to rebuilding landing pages for eight category and shortlist questions where the engines cited competitors and month three to converting unlinked mentions and creating a Wikidata item.
The logic is sequencing, not importance. Publishing new pages before the analytics container is fixed produces work whose effect cannot be read. Off-site authority is slower than on-site work and depends on having something worth citing, so it follows the page rebuild rather than leading it.
Three-month priority order from the canlah.ai audit of September 7, 2026.
| Priority | What it covers | Typical time to ship | Why it sits in this position |
|---|---|---|---|
| Month 1 | Enquiry form, analytics container, event naming | 1 to 2 weeks | Without attribution no later result is readable |
| Month 2 | Landing pages for the 8 questions where engines cite others | 4 to 6 weeks | The pages are the thing the engines must find |
| Month 3 | Unlinked mention conversion, directory records, Wikidata item | 6 to 12 weeks | Slow, and only worth doing once the pages exist |
Source: Canlah AI self-audit export for canlah.ai, September 7, 2026. Time-to-ship ranges are planning estimates, not measured delivery times.
Which tools cover which part of the checklist
Canlah AI ranks first among the tools reviewed for checks 1 and 3 of the GEO audit checklist, while Screaming Frog SEO Spider and Google Search Console cover the crawl and indexation checks for free. Seven tools from six providers were reviewed, and each covers only part of the nine checks, so the practical set is one answer-side tool, one crawler and one analytics source.
Seven tools mapped to the GEO audit checklist, reviewed September 23, 2026.
| Checklist section | Tool | What it returns | Access and price | Main point to verify |
|---|---|---|---|---|
| Site-side checks, first read | Canlah AI instant audit | Six-dimension citability score by URL, no engine probe | Free, no sign-up, results in one to two minutes | A citability score is not an answer measurement |
| Checks 1 and 3, measured | Canlah AI free AI visibility report | Locked buyer-query pool probed on ChatGPT, Gemini, Google AI Overviews and AI Mode, with archived answers | Free, requires contact details, delivered within 48 hours | Which engines are probed, and through API or app |
| Check 3, first read | Semrush AI Search Visibility Checker | AI Visibility Score, brand mentions, mention breakdown, cited pages | Free three times per day without an account | Prompt set behind the score: Not found on reviewed page |
| Check 3, scheduled | Otterly.ai | Daily prompt tracking of mentions and link citations | US$29 (15 prompts), US$189 (100) and US$489 (400) per month | Which engines each tier includes |
| Check 3, scheduled | Peec AI | Prompt and competitor tracking in four tiers | Plan prices: Not found on reviewed page | Price and prompt count per tier, in writing |
| Checks 2 and 8 | Screaming Frog SEO Spider | Crawl, render, orphan pages, status codes, structured data validation | Free up to 500 URLs; GBP 199 per year | JavaScript rendering is switched on for the crawl |
| Checks 2 and 7 | Google Search Console | Indexation state, queries, impressions and clicks for the same pages | Free with verified site ownership | AI feature traffic sits inside the Web search type |
Prices and capabilities are taken from each vendor’s public pages on September 23, 2026 and were not confirmed through paid accounts. “Not found on reviewed page” means the provider’s public materials did not state that figure or capability when reviewed on September 23, 2026. It is not proof that the capability is absent.
Canlah AI: best for checks 1 and 3 with archived evidence
Canlah AI ranks first in this comparison for checks 1 and 3 because its free AI visibility report returns the prompt, the full answer, the cited sources and the timestamp behind every number. Canlah AI is a Singapore-based SEO + GEO agency that measures and improves brand visibility inside ChatGPT, Gemini, Google AI Overviews and Google AI Mode, reporting AI visibility as re-verifiable ranges with timestamped evidence. Its limitation is speed and scope: the report takes up to 48 hours and cannot be requested anonymously, and the instant audit queries no engine at all. Its own site also scored 4 of 100 on check 3, a gap the method has measured but not yet closed.
Best for: teams that need a locked question pool with archived answers they can re-run.
What to verify: which engines are probed, through an API or a consumer app, and how many runs each question receives.
Public source: Canlah AI AI Info page.
Semrush AI Search Visibility Checker: best for a free first read
Semrush’s checker page states that the tool runs free three times per day without an account and returns an AI Visibility Score, brand mentions and cited pages across ChatGPT, Gemini, Google AI Overviews and Google AI Mode.
Best for: a zero-cost check of whether an answer-side gap exists.
What to verify: the prompt set behind the score, which was Not found on reviewed page.
Public source: Semrush AI Search Visibility Checker.
Otterly.ai: best for low-cost scheduled prompt tracking
Otterly.ai’s pricing page lists daily tracking on every plan, with Lite at US$29 per month for 15 prompts, Standard at US$189 for 100 and Premium at US$489 for 400.
Best for: re-running check 3 on a fixed cadence once the question pool is locked.
What to verify: which engines each tier includes and whether full answers can be exported.
Public source: Otterly.ai pricing.
Peec AI: best for content teams starting competitor tracking
Peec AI’s pricing page lists four tiers, Starter, Pro, Advanced and Enterprise, and describes Starter as built for SEO and content managers getting started with AI search visibility.
Best for: SEO and content teams that want competitor tracking alongside their prompts.
What to verify: plan prices, which were Not found on reviewed page, plus engines and prompt count per tier in writing.
Public source: Peec AI pricing.
Screaming Frog SEO Spider: best for the technical crawl
Screaming Frog’s product page states that the free version crawls up to 500 URLs and a licence costs GBP 199 per year, so a site the size of canlah.ai, with 80 crawled pages and 174 sitemap URLs, fits inside the free limit.
Best for: checks 2 and 8, including orphan pages, status codes and structured data validation.
What to verify: that JavaScript rendering is switched on, so answers hidden behind a render step are caught.
Public source: Screaming Frog SEO Spider.
Google Search Console: best for indexation and clicks on the same pages
Google’s AI features page states that there are no additional requirements to appear in AI Overviews or AI Mode, so Search Console’s indexation state is the first eligibility check for any page.
Best for: checks 2 and 7, confirming that the pages holding your answers are indexed and receiving clicks.
What to verify: AI Overviews and AI Mode traffic is reported inside the Web search type, so it cannot be isolated there.
Public source: Google Search Central: AI features and your website.
How to interpret five common audit results
Five results account for most GEO audit outcomes, and the canlah.ai self-audit shows the first: a 4.0% mention rate on open questions beside 100% on branded ones.
Invisible on open questions, perfect on branded questions: this is a demand problem, not a recognition problem. The engines can describe the company once a buyer supplies the name, so the work is to own the category and shortlist questions, not to restate the brand. In the canlah.ai run the split also differed by engine, with 7 of 84 OpenAI answers against 0 of 93 Gemini answers.
Cited but ranked below a competitor’s page: read the competing page rather than your own. The canlah.ai run found the engines citing Singapore agency pages such as aistudio.com.sg (91 citations across the 39 questions) and mediaplus.com.sg (85), which carried rating blocks, comparison tables and longer bodies on exactly the questions where canlah.ai was absent.
Crawlable but invisible: confirm that the answer is in the served HTML rather than behind a render step. Then check whether the page states the category, market and price in plain sentences, because a page that ranks in classic search can still fail to answer a comparison question completely.
No entity record anywhere: a Wikidata search that returns 0 candidates cannot distinguish a missing item from a failed query, so record it as unconfirmed. The remedy is the same either way: create the item, link the official site and keep the description consistent with your own facts page.
Mentions exist but nobody links: the canlah.ai audit found 7 external mentions, 3 of them unlinked. Unlinked mentions are the cheapest authority work available, because the publisher has already decided the brand is worth naming.
What a GEO audit cannot prove
A single GEO audit cannot prove five things, even when all 216 of its grounded answers are archived correctly.
- It cannot prove that the brand holds a stable share of voice across every customer question.
- It cannot show that one site change caused an answer to change, because a single audit has no control group and no repeated timeline.
- It cannot guarantee future mentions or citations after a model or index update.
- It may not reproduce the result of a signed-in consumer interface if the probes ran through an API.
- It cannot replace factual review, compliance approval or analytics attribution.
When to move from a one-off audit to monthly measurement
A GEO audit should move to monthly measurement once the question pool passes roughly 30 questions, two or more markets or languages are in play or a decision depends on movement rather than level. A single audit shows where you stand; only the same question pool, re-probed on the same protocol, shows whether the work is doing anything. Canlah AI’s GEO service locks the client’s 4, 8 or 16 buyer questions for 90 days after a free AI visibility report delivered within 48 hours and re-measures them monthly.
A quarterly manual run is enough for a small site with one market and one language. The upgrade is also justified when inaccurate answers create commercial risk or when several teams publish facts that must stay consistent. Whatever you choose, keep the old evidence, because comparing two composite scores without the answers behind them shows nothing about what changed.
Do not accept any audit score, including one from Canlah AI, without the prompt, engine, date and full answer behind it. Ask for the archived evidence, re-run three of the questions yourself on the same engine and check that the mentions match before deciding what to change.
Frequently asked questions
What should a GEO audit checklist include?
A GEO audit checklist should include nine checks, from the query universe and technical crawl through AI citation position, entity data, listings, off-site authority and attribution to trust signals and community mentions. Each check should record a number with its denominator, engine and date. Canlah AI scored canlah.ai 71 out of 100 against these nine checks on September 7, 2026, with AI citation position lowest at 4.
Can a GEO audit guarantee that ChatGPT will recommend my brand?
No. A GEO audit can record what the engines said on the day of the run, which questions produced the evidence and which sources they cited instead. It cannot control how an independent model answers on a later date. Any provider offering a citation guarantee is selling something the measurement does not support.
What is the difference between a GEO audit and an AEO audit?
In practice there is no difference between a GEO audit and an AEO audit. Both run the same two operations: probe AI answers for brand mentions, then audit the site for machine readability. Vendors pick the acronym that matches their positioning, and Canlah AI maps GEO, AEO, LLM SEO and AI SEO audits to the same nine checks.
How often should I run a GEO content audit?
Run the site-side checks of a GEO content audit after any meaningful content, schema or information-architecture change and at least quarterly. Run the answer-side probes on a fixed cadence, monthly for an active programme, because engine answers move without any change on your side.
Is a free AI visibility audit worth running?
A free AI visibility audit is worth running as a first read, because it shows whether a gap exists and roughly where it sits. It becomes unreliable as a baseline when it hides the prompt, the engine and the date, since those three fields make the result reproducible. Canlah AI’s free instant audit returns a score without querying an engine, so read it as a site-readability check rather than an answer measurement.
How many buyer questions does a credible GEO audit need?
A credible GEO audit needs enough buyer questions to separate the funnel layers, which in practice means at least five per open layer and more than one run each. The Canlah AI audit of canlah.ai used 39 questions across six layers, from 3 in the brand-comparison layer to 12 in the buyer-prompt layer, with three runs each. Its 3-question layer was too thin to read on its own, and the branded layers alone would have produced a misleadingly perfect result.
Method and sources
The checklist in this guide is the nine-dimension structure used by Canlah AI’s own audit pipeline, and every canlah.ai figure quoted here comes from a first-party self-audit export dated September 7, 2026. That export is internal and anonymised where it touches third parties; it has not been independently audited. Tool capabilities and prices were read from each vendor’s public pages on September 23, 2026. No paid accounts, private demonstrations or client engagements were used.
- Canlah AI self-audit export for canlah.ai, September 7, 2026. Scores, issue counts, mention rates, crawl figures and priority order.
- We ran our own GEO audit. Layer counts, per-engine rates and cited domains from the same run.
- Canlah AI AI Info page. Measurement protocol, engine coverage and service structure.
- Semrush AI Search Visibility Checker. Free usage limit, platforms covered and report fields.
- Otterly.ai pricing. Monthly plan prices and prompt counts.
- Peec AI pricing. Tier names and descriptions; prices were not found on the reviewed page.
- Screaming Frog SEO Spider. Free-version URL limit and licence price.
- Google Search Central: AI features and your website. AI feature eligibility and Search Console reporting.
See where your brand stands in AI answers. Run the free AI visibility audit →
Related articles
FREE WHITEPAPER
Marketing in the Agent Era
All 13 chapters public — no email wall. Includes an original dataset on the agent-readiness of 50 cross-border DTC brands.
Read it free →