How Spyglo scores your website
Spyglo runs 43 checks on a website. 20 of them make up two scores out of 100: an SEO score from 12 checks and a GEO score from 8. The overall score is the average of the two. The other 23 checks show up in your report as issues to fix, but don’t move a score. This page lists them all, with the exact weights the scanner uses.
What Spyglo reads
The page you enter (usually your homepage), your robots.txt file, your sitemap and your llms.txt file, fetched live, as a search engine or AI crawler would see them. Spyglo doesn’t run your page’s JavaScript, because the main AI crawlers don’t either. It only reads pages anyone can see and never logs in or installs anything.
The SEO score
Whether search engines can find, read and understand the page. Each check is worth a share of 100; pass it and you get its share.
| Check | Share of the score |
|---|---|
| No duplicate title/meta-description/canonical tags | 11% |
| Title tag present, 10-70 characters | 9% |
| Meta description present, 50-165 characters | 9% |
| Exactly one H1 | 9% |
| Served over HTTPS | 9% |
| robots.txt present and not blocking all crawlers | 9% |
| Sitemap present with URLs | 9% |
| At least 80% of images have alt text | 9% |
| Canonical tag present | 7% |
| Viewport meta tag present | 7% |
| www / non-www consolidated to one host | 7% |
| HTML lang attribute set | 5% |
The GEO score
Whether AI assistants like ChatGPT, Claude, Perplexity and Google’s AI Overviews can reach the page, understand who you are and find something worth quoting.
| Check | Share of the score |
|---|---|
| AI search bots can reach the site (ChatGPT, Claude, Perplexity, Google, Bing) | 23% |
| Entity or author schema present | 16% |
| Content includes quantified, checkable claims | 14% |
| Structured data (JSON-LD) present | 11% |
| At least 300 words of content | 11% |
| At least one question-phrased heading | 9% |
| Sitemap carries lastmod freshness dates | 9% |
| FAQ / answer schema present | 7% |
Shares are rounded, so a column may not add up to exactly 100.
Why the GEO weights are what they are
- Letting AI search bots in carries the most weight. If a bot can’t reach your site, nothing else matters to that assistant.
- Facts, question headings and clear identity come next. Adding sources, quotations and statistics to a page made it up to 40% more visible in AI-generated answers. (Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024). Schema that says who you are helps machines trust and attribute what they read.
- FAQ schema is a small bonus, and llms.txt no longer counts. Google says its AI features need no special files or schema (Google Search Central), and it stopped showing FAQ rich results in May 2026. We still tell you whether you have an llms.txt file, but it doesn’t move your score.
- Link preview tags don’t count towards SEO. Open Graph and Twitter/X tags make shared links look good, but Google doesn’t use them to rank pages, so we report them without scoring them.
Which AI bots we check
Spyglo reads your robots.txt the way the bots do (the RFC 9309 rules: the most specific group, the longest matching rule, Allow wins a tie). Blocking a search bot keeps you out of that assistant’s answers, so it costs GEO points. Blocking a training bot is your choice and costs nothing.
Search and citation bots (blocking these lowers your GEO score)
- OAI-SearchBot: ChatGPT search
- ChatGPT-User: ChatGPT, when someone asks it to read a page
- Claude-SearchBot: Claude search
- Claude-User: Claude, when someone asks it to read a page
- PerplexityBot: Perplexity
- Perplexity-User: Perplexity, when someone asks it to read a page
- Googlebot: Google Search and AI Overviews
- Bingbot: Bing and Microsoft Copilot
Training bots (reported, never scored)
- GPTBot: OpenAI model training
- ClaudeBot: Anthropic model training
- Google-Extended: Gemini training
- Applebot-Extended: Apple model training
- Meta-ExternalAgent: Meta AI training
- CCBot: Common Crawl, used for training
See exactly what your own robots.txt allows with the free AI Crawler Checker.
Checks that are reported but not scored
These catch real problems, but either overlap with a scored check or matter on some sites and not others, so they show as issues without changing your score:
- An llms.txt file (reported only: Google says it isn’t needed)
- Twitter/X card tags, for link previews
- The page isn’t set to noindex
- No meta refresh redirect
- The words can be read without running JavaScript
- The canonical tag has no protocol, domain or parameter problems
- The viewport doesn’t stop people zooming
- The viewport is set up for phones
- Images have a width and height set
- The homepage is listed in the sitemap
- robots.txt points to the sitemap
- The page links out to trusted sources
- The page uses main or article landmarks
- Sitemap and canonical addresses don’t redirect
- og:url is the page’s own address
- Open Graph title and description, for link previews
- Your server and firewall let AI search bots in, not just robots.txt
- Cloudflare Content Signals don’t ask AI to stay out of answers
- AI agents: buttons and links have names they can read
- AI agents: form fields are labelled
- AI agents: a clear way to contact or book
- AI agents: no CAPTCHA on the homepage
- AI agents: contact or buying details in your schema
What Spyglo doesn’t do
- It doesn’t claim to know where you rank inside ChatGPT. AI answers change almost every time, so the free AI Visibility Test helps you ask the assistants yourself instead.
- It doesn’t crawl every page of your site in the free check; it checks the page you enter plus your site-wide files.
- It can’t promise a ranking. Nobody can. It shows what’s in the way.
Change log
- 7 November 2026: scores cleaned up to match what Google and the AI assistants really use: llms.txt and the Open Graph and Twitter/X tags are now reported but not scored, and question headings count for less. New checks: whether your server or firewall (often Cloudflare) turns away AI search bots even when robots.txt lets them in, Cloudflare’s Content Signals, and five “Can AI agents use your site?” checks, which have their own verdict and don’t change your scores.
- 26 October 2026: two new checks, reported but not scored: sitemap and canonical addresses that redirect (Google won’t index them and lists them as “Page with redirect”), and an og:url that names a different page. Reports now explain that after a www redirect, Search Console lists the old address as “Page with redirect”, which is expected.
- 26 October 2026: this page now reads every check and weight straight from the scanner, and shows all 43 checks.
- 14 October 2026: landmarks pass with either a main or an article element; any organisation type in schema (Dentist, Plumber, RoofingContractor and so on) counts as entity schema.
- 8 October 2026: added the “words readable without JavaScript” check and Perplexity-User to the citation bots.
- 4 October 2026: GEO weights rebalanced: FAQ schema and llms.txt down to small bonuses, AI bot access, entity schema, question headings and facts up. An empty Disallow line no longer counts as blocking.
Run a free check to see how your site does on every one of these.