GEO Score Card · updated October 2026

Is AI recommending you, or your competitors?

ChatGPT, Gemini, Perplexity and Google AI Mode build shortlists before your buyer ever clicks. Score your site in 10 minutes, then let us show you what the engines actually say about you.

Score yourself in 10 minutes Request a free AI Visibility Snapshot

2009

doing SEO since

241

weekly SEOs Diners Club issues

Hundreds

of sites run through this card

Snapshot excerpt · anonymized

Real audit

Buyer question asked across five engines:

“people counting sensor for retail stores with POS integration”

Google AI ModeNamed · 1st

PerplexityNamed · 1st

ChatGPTNamed · 3rd

Their enterprise customers0 of 6 answers

They win the category question, but every engine files them as the small-store option. That is the kind of finding a checklist can't give you.

How AI decides who to recommend

Every brand that gets recommended walks through three doors.

We built this model after running hundreds of sites through the Score Card. Most fail at least one door without knowing it. Rankings get you into the pool. The doors decide whether you get named.

Reach

Can the engines get to your pages, and read the first 200 characters that actually matter?

About 3 in 4
sites we scored give AI search bots no clear access.

Recognize

Does the model know you as an entity, in the right category, next to the right competitors?

About 87%
of sites we scored have no Wikidata entry.

Believe

Is there independent proof the engine can quote, or only what you say about yourself?

About half
have no independent proof an engine can cite.

The Score Card · 38 checks · about 10 minutes

Tick what is already true for your site.

Each check carries the reason it matters. Your score updates as you go. Nothing is sent until you ask for your results.

Reach

Can AI search engines get to your pages and read what matters?

AI search bots are allowed: OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBotCritical+10

Training bots like GPTBot are a separate decision. Many sites block all of them with one wildcard rule.

Your CDN or firewall is not blocking AI crawlers by defaultCritical+10

Cloudflare and some WAF presets block AI bots unless you opt in. robots.txt can say yes while the firewall says no.

Key pages are indexable in Google and BingCritical+10

No stray noindex, canonicals pointing to the right URL. Engines ground answers in search indexes.

The core answer is in plain HTML, not drawn by JavaScriptCritical+10

Most AI crawlers do not render JavaScript. If the answer only appears after scripts run, they don't see it.

Every key page has one H1 and a clean first 200 charactersHigh+7

Engines keep roughly the title, the URL and about 200 characters of body. Dates, breadcrumbs and image alt text eat that budget.

Top pages open with the answer, not a warm-upHigh+7

Put the one-sentence answer first. Citations come mostly from the top of the page.

Snippet controls don't block AI featuresHigh+7

nosnippet or a tight max-snippet can keep your page out of AI answers.

Your top 5 pages are in ChatGPT's cache, and recentMedium+4

Free test: ask about them in ChatGPT's lockdown mode, which can only show cached pages. Note the version date.

XML sitemap submitted to Google and Bing, IndexNow onMedium+4

Faster discovery of new and updated pages.

Core Web Vitals pass: LCP under 2.5s, INP under 200ms, CLS under 0.1Medium+4

Slow pages are expensive to fetch live, so engines lean on older cached copies.

PageSpeed Insights Agentic Browsing checks reviewedLow+2

The accessibility tree is how AI agents read and use your page. Early signal, not a ranking factor.

llms.txt published (optional)Low+2

Low cost, low impact. Most llms.txt files get almost no requests, and Google does not use it for ranking.

Recognize

Does the model know who you are, and put you in the right category?

Asked who your competitors are, AI names the right onesCritical+10

The competitor list is the mirror of the category the model filed you in. Wrong list, wrong positioning.

One plain sentence defines your category on the homepage and About pageCritical+10

Category plus brand, no adjectives. The same sentence everywhere.

Organization schema with @id, logo, founder and sameAs linksCritical+10

Ties your site to your profiles and your Wikidata entry so engines see one entity.

Named in a best-in-category answer with web search turned offHigh+7

Run it 5 times in new chats. Brands named from memory get mentioned far more often than brands only found by search.

Wikidata entry with aliases and referenced factsHigh+7

About 87% of sites we scored have none. It is the main fix when your name gets confused with another brand.

Name, address, phone and boilerplate identical on every profileHigh+7

Small variations split one brand into several weak entities.

Real author pages with Person schemaHigh+7

Named experts with a track record are easier to trust and to cite.

About page answers who, what, since when and for whomHigh+7

The page engines check first when they verify a claim about you.

No AI hallucinations about your brand, or their source is tracedHigh+7

Fixing your own page is rarely enough. Find the third-party page feeding the error.

Your name isn't confused with a similar-sounding entityMedium+4

Test with search on and off. Confusion usually lives in the model's memory, not in search.

LinkedIn company and founder profiles match the siteMedium+4

Same category line, same founding year, same links.

Google Business Profile claimed and complete (if you serve locally)Medium+4

Local answers draw on business profiles and review platforms.

Believe

Is there independent proof an engine can quote?

Independent proof is documented on your site: awards, rankings, reviewsCritical+10

Proof beats posture. The page that gets cited documents someone else's praise, not its own.

You appear in the third-party lists and comparisons engines cite for your categoryCritical+10

Your own site is a small share of AI sources. Category answers are assembled from other people's pages.

Customer outcomes are named in text, not only shown as logosHigh+7

An engine cannot quote a logo.

Active review profiles: G2, Capterra, Trustpilot or your vertical's equivalentHigh+7

Review platforms are a common source for best-of answers.

Original data or research published on your own domainHigh+7

First-hand numbers are what other sites quote, and what engines cite.

Statistics, quotes and sources cited inside your contentHigh+7

The GEO study (Aggarwal et al., 2024) found these lift visibility in AI answers by up to 40%.

Visible dates and a real content refresh every quarterMedium+4

Change the substance, not just the date.

No self-ranked best-of lists that put you at number oneMedium+4

They may get cited, but they rarely get you recommended.

Your brand is discussed in communities: Reddit, YouTube, forumsMedium+4

Authentic participation only. Platform shares move fast, so measure rather than assume.

Earned media mentions in the last 12 monthsMedium+4

Independent coverage feeds the consensus engines rely on.

Measure

Can you prove what moved?

A monthly prompt panel: same prompts, 4+ engines, model and date loggedCritical+10

Answers change with every model update. One run is a lead, not a finding.

Search Console's generative AI performance report reviewed monthlyHigh+7

Open to all sites since August 31, 2026. Impressions only, no clicks.

AI referrals tracked in GA4 (chatgpt.com, perplexity.ai, gemini.google.com)High+7

Separate them from other referrals so you can see the trend.

Bing AI Performance and Microsoft Clarity Citations connectedMedium+4

Free first-party citation data from the Microsoft side.

Your level

Invisible

Start with the Reach door. Nothing else holds until engines can get in.

Reach0%

Recognize0%

Believe0%

Measure0%

Email me my results

You get your score, your weakest door and the three gaps we would close first.

Sent · check your inbox

Your weakest door:

The three gaps we would close first:

Free AI Visibility Snapshot

A checklist shows what is in place. A Snapshot shows what the engines say.

One page, written by hand. We ask your buyers' real questions across five engines, record who gets named instead of you, and rank three gaps in dependency order, because the second fix doesn't hold until the first one is closed.

Google AI Mode, ChatGPT, Perplexity, Claude and Gemini, logged out, with run dates
The competitors who were named, and which sources the engines cited for them
Three gaps in dependency order, each with an honest effort tag
What a paid baseline would add: full crawl, bot-access test, 25 prompts at three runs each

Score yourself, then request it8 to 10 a month · best fit: teams without an SEO agency on retainer

AI Visibility Snapshot

Retail analytics company

Anonymized
September 2026

5 of 5

engines named them

4 of 5

put them first

0 of 6

answers named a customer

Three gaps, in dependency order

Every engine files them as the small-store choice

Effort: one afternoon

Their best-known customers appear in no answer

Effort: under a week

Another vendor wrote the sentence about their product

Effort: a quarter, ongoing

Three steps, each useful on its own

Measure, fix in the right order, then prove it moved.

Step 1 · Free · instant

Score yourself

38 checks across three doors. Your results and top gaps by email.

Start the Score Card

Step 2 · Free · 8 to 10 a month

AI Visibility Snapshot

Five engines, real buyer questions, three gaps in dependency order. One page, by hand.

See what's inside

Step 3 · Paid

Baseline and program

25 prompts, four engines, three runs each. We produce the fixes, your team ships them, and week 12 is an exit gate: same prompts, side by side.

Talk to us

From SEOs Diners Club

GEO notes, measured and dated.

Every check above traces back to something we tested and wrote up. Eight of the notes behind this card:

Subscribe, free

Access#240 · Sep 28

Four bots, four decisions

Block training if you want. Keep search, user and ads bots open.

Access#237 · Sep 7

ChatGPT sees about 200 characters of you

Its own index family, a cache layer, and a free test for both.

Entity#234 · Aug 17

Known beats findable: 68.9% vs 2.1%

ChatGPT writes its shortlist from memory before it searches.

Research#233 · Aug 10

Hundreds of sites, three doors

Reach, Recognize, Believe: where most sites fail.

Entity#232 · Aug 3

The goal is the citation, not the backlink

Audit the source gap against the competitors who get named.

Research#230 · Jul 20

No local G2? Ads fill the answer

10 sectors, 2 engines: what AI does when review sites are missing.

Content#229 · Jul 13

Six GEO myths, and proof beats posture

Training data vs grounding, and why one award page got cited.

Entity#226 · Jun 22

Cited is not recommended

Calling yourself the best hands the win to competitors.

Questions we get

Is the Score Card enough to know how AI sees us?

No. It tells you what is in place on your side. It can't tell you what ChatGPT or Gemini actually say about you, or who they name instead. That is what the Snapshot shows.

Why only 8 to 10 Snapshots a month?

Each one is run and written by hand: real prompts, five engines, logged out, with dates. There is no auto-generated report behind it.

Do you guarantee AI citations?

No one honestly can. Answers are personalized and the default models change every few weeks. We measure, fix in dependency order and re-run the same prompts to show what moved.

Is GEO different from SEO?

Good SEO is most of good GEO. Rankings get you into the pool engines read from. GEO adds what decides who gets named: entity clarity, independent proof and access for AI search bots.

Who is the Snapshot a fit for?

Companies selling in the US, UK or Australia that don't already have an SEO agency on retainer. If you do, the Score Card results are still yours to keep.

Mert Erkal

Founder, Stradiji

Doing SEO since 2009, founded Stradiji in Istanbul in 2010. Writes SEOs Diners Club every week and tests every GEO claim live before it lands on this card.

Last updated · October 7, 2026

What changed: rebuilt around three doors; added cache, memory and bucket tests; split AI search bots from training bots; removed ai.txt, AGENTS.md and SKILL.md; llms.txt downgraded to optional.

Find out who AI names when your buyer asks.

Score yourself Get a Snapshot