Beyond HTML audits: the 4 GEO signals I actually watch for in 2026
Saw the Claude in Chrome GEO audit prompt going around last week. Good starting point — the raw HTML vs rendered DOM separation is smart. But after months of running GEO audits, I've found 4 dimensions HTML-only audits systematically miss.
I call it the SEEC framework — Sources, Entities, Evidence, Consistency.
1. Sources — beyond robots.txt
Most audits check robots.txt for GPTBot / ClaudeBot / PerplexityBot. That's necessary but not sufficient. In 2026 you should also check:
llms.txtat root — still emerging but some AI agents fetch it when present. Curated table of contents for LLMs.- RSS / Atom feeds — many AI crawlers still discover fresh content via feeds.
- Sitemap freshness —
<lastmod>accuracy matters more than sitemap existence.
Quick test: curl -A "GPTBot" https://yoursite.com/llms.txt. If you don't have one, your competitors probably don't either. Window of opportunity.
2. Entities — the Wikidata / Wikipedia signal
LLMs anchor facts to entities. Brands with a stable Wikidata QID + Wikipedia article get cited disproportionately more in ChatGPT / Perplexity / Gemini answers.
Audit checklist:
- Does the brand have a Wikidata QID?
- Is it referenced via
sameAsin the Organization JSON-LD? - Are Wikipedia mentions consistent with the schema data?
- Does the founder / CEO have their own Person entity, also referenced?
3. Evidence — citability of your content
The Claude prompt checks for "stats, dates, sources, author attribution". Right idea, but granularity is off. What actually matters:
- Specific numbers with sources — "45% of…" beats "many companies…"
- Dated statements — "As of Q1 2026" beats undated claims
- Named authors with credentials — with matching Person schema
- Direct quotes with attribution — LLMs love these
Rule of thumb: if a paragraph can't be cited standalone in an AI answer, it won't be.
4. Consistency — the cross-source check
The one nobody audits. LLMs cross-reference facts across sources. If your Google Business Profile says one thing, Wikipedia another, your schema a third — LLMs don't cite conflicting entities.
Manual check: pick 5 facts from your homepage (founding year, HQ location, employee count, main service, key metric). Search each on Google, Perplexity, ChatGPT. Do all sources agree? If not, that's your GEO fix list.
What I don't audit (and neither should you, probably)
- Word count — irrelevant for GEO, concise beats bloated
- Keyword density — dead concept
- Site speed specifically for GEO — CWV still matter for classic SEO, much less for AI answer inclusion
Curious what signals others are tracking. What did I miss?