Quick answer: SeeGeo's audit crawls your site once, politely, the way an AI crawler does — no JavaScript execution — and scores it across six categories: crawler access, technical foundation, structured data, content extractability, off-site presence, and entity clarity. Categories combine into two pillar scores (classic SEO readiness and GEO — readiness to be cited in AI answers), which average into a 0–100 number and a letter grade. The letters are anchored to reality: an A means the top tenth of the 107 real small-business websites we measured in our published study, not an arbitrary cutoff. The score is deterministic — same site in, same score out — every report states exactly what was and wasn't evaluated, and a provider outage can never read as your invisibility. The rest of this post is the whole rubric, including the parts that make grades go down, the parts that deliberately can't, and why an F for google.com is both correct and meaningless.
Scoring tools usually treat their formula as a trade secret. We think that's backwards: a score you can't interrogate is a score you can't trust, and trust is the entire product. So here's ours.
The crawl: we read your site the way AI does
Every audit starts with one polite crawl — your robots.txt, your sitemap, your homepage, and a budgeted set of key pages, fetched with our honest user-agent and without executing JavaScript. That last choice is the load-bearing one: Vercel and MERJ measured that the major AI crawlers download JavaScript but don't run it, so a page whose content only exists after scripts run is, to most AI systems, an empty room. We deliberately see what they see. If your content survives our crawl, it survives theirs.
The six categories
1. Crawlability & access — the gate. Can crawlers get in at all? We evaluate your robots.txt against sixteen named crawlers individually — classic search bots and the AI roster (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended, and friends) — plus bot-walls, noindex directives, and whether your security layer answers instead of your site. This category is special: a critical failure here doesn't just lower a number, it caps your whole grade, because nothing else matters if machines can't read you. The cap is proportional — blocking one minor AI crawler costs you far less than walling out everything — and if the blocking looks deliberate (the pattern big publishers use on purpose), the report says so instead of scolding you.
2. Technical foundation. Page speed, mobile viewport, HTTPS hygiene, clean URLs, sitemap health. Classic, unglamorous, still counted.
3. Structured data. The schema.org markup that lets a machine assert facts about you instead of inferring them — LocalBusiness, Organization, FAQ, Product. In our study, 27.1% of real small-business sites had none at all.
4. Content extractability. Could an AI quote you? Clear headings, real text (not text baked into images), answerable questions, statistics and sources. This is where the Princeton GEO research says most citation variance lives, so it carries the most GEO weight.
5. Off-site presence. Wikipedia and mention analysis — does the wider web know you exist? Honesty note that matters: this category's data sources are still growing, so its weight in your score scales with how many sources actually answered. We'd rather a category count for less than pretend one lookup is a full picture.
6. Entity clarity. Can a machine state, with confidence, what you are, where you operate, and for whom? Our classifier tries — and if it can't classify you, neither can an LLM. This was the most-failed check in our study: 64.5% of sites never plainly say what the business does at the top of the homepage. It's also the cheapest thing on this list to fix.
You'll notice what's not a category: which AI engines you've paid to appear in (not a thing that exists), your follower counts, or a secret sauce coefficient. Volume of honest signal, nothing else.
Two pillars, one grade — anchored to real sites
Each category feeds two pillar scores with different weights: SEO (search readiness — access and technical matter most) and GEO (AI-answer readiness — extractability and off-site matter most). The pillars average into your combined score, and the combined score becomes a letter.
Here's the part we changed recently, and why your letter is defensible in a way most tools' letters aren't: the grade cuts come from the score distribution of 107 real small-business websites — restaurants, trades, clinics, shops across seven countries — from our published study. An A means the top decile of real sites. B is the top quartile. C clears the median. Under our old arbitrary cutoffs, the best site we ever measured graded B and nothing could earn an A — a ruler nobody could calibrate against. Now the letter answers a real question: compared to actual businesses competing for the same AI answers, where do you stand?
The rules that protect you from us
A scoring system reveals its character in its failure modes. Ours has four rules that all bend the same direction:
An error can never read as invisibility. If Wikipedia times out mid-audit, we retry; if it's truly unreachable, the report says "source unanswered" — it never says "no Wikipedia presence" because our network hiccuped. The same rule runs through our tracking product: failed API calls are flagged and excluded from every statistic.
Unevaluated never counts against you. If a category couldn't be measured, it's removed and the weights renormalize — and the report says so, right under the grade: "Scored on 5 of 6 categories", with the reason. A screenshot of your grade carries its own caveat.
Same site in, same score out. The score is deterministic — no LLM judgment folded into the number, ever. When we do show you AI-generated readings (the "How an AI reads your site" panel, where a language model reads your homepage exactly as a crawler sees it), they're labeled, and they sit beside the score, never inside it. A tracked score only means something if the ruler holds still.
The ruler is versioned. When we improve the engine — new checks, re-anchored grades — the version number on your report changes and the change is documented. If your grade moves between audits, you can tell whether your site changed or our measurement did. We built this after watching a site grade differently two days apart and realizing we couldn't prove which had moved.
So why does google.com get an F?
Because the audit measures one thing: whether a machine that has never heard of a business can learn what it is from its website. Google.com scores a perfect 100 on crawler access — and an F overall, because its homepage is a search box: no self-description, no structured data about the business, nothing to extract. Every one of those findings is true. And every one is meaningless as a verdict on Google, because household names live in every AI model's training data and get recommended no matter what their homepage says.
That's exactly the problem an unknown business doesn't have the luxury of. If you're not famous, the machines learn who you are from your website or they don't learn it at all — which is who this rubric is built for. (The report says all of this in context when you audit a famous site, before showing the grade. Skeptics testing us with google.com first: we see you, and fair enough.)
What the score doesn't claim
The audit is a measurement of readiness, not a promise of outcomes. GEO is a young discipline; our recommendations reflect current research and practitioner consensus, not settled science, and every report says so verbatim. The audit also can't see whether AI engines actually mention you today — that's what tracking measures, with repeated live queries across engines, because single answers are noise and trends are signal.
Run it yourself — it's free, about 30 seconds, score with no signup. And if you disagree with a finding, the evidence is printed under it: the robots.txt line, the missing markup, the page we fetched. Argue with the evidence, not with a black box. That's the point.
Frequently asked questions
How does SeeGeo calculate its website score? One JavaScript-free crawl feeds six category scores (crawler access, technical, structured data, extractability, off-site presence, entity clarity), which combine into SEO and GEO pillar scores and average into a 0–100 number. Letter grades are percentile-anchored: A = the top tenth of the 107 real small-business sites in our published study.
Why did a famous website score badly on your audit? Because the audit measures whether a machine can learn what a business is from its website — the problem unknown businesses have and famous ones don't. A household name's homepage is often a pure application with no self-description, which grades honestly low while meaning nothing about the brand's actual AI visibility. Famous-site reports say this in context before the grade.
Is the score affected by AI randomness? No. The audit score is fully deterministic — same site in, same score out — with no LLM judgment inside the number. AI-generated readings appear beside the score, clearly labeled. Our tracking product handles AI answer randomness separately, by measuring across repeated runs.
What does "scored on 5 of 6 categories" mean? A category that couldn't be evaluated — say, an off-site data source didn't answer — is excluded and the remaining weights renormalize, so a missing data source never punishes your site. The report states which category was skipped and why.
Can I improve my grade, and how fast? Usually, yes — the report is a prioritized fix list, impact-first, and the most common failures (no plain-language identity statement, no structured data) are cheap to fix. Re-audit any time for free; the engine version on each report tells you the comparison is apples-to-apples.