ChatGPT does not use PageRank. It does not crawl anchor text. It does not reward domain authority in the way Google does. If your ChatGPT SEO strategy is a subset of your Google strategy, you are optimising for a different system than the one you think you are. The research has now accumulated enough to say what actually works.
Why traditional SEO fails as a ChatGPT strategy
A peer-reviewed arXiv study on AI source selection found that GPT-4o draws from a source pool that overlaps with Google's indexed results by just 4%. That means 96% of what ChatGPT cites is outside the territory traditional SEO is fighting for.
This is not incidental. ChatGPT is trained on a massive corpus of web text and selects sources through retrieval mechanisms that weight different signals than PageRank. Backlinks — the core authority signal in Google's algorithm — are a structural relationship between web pages that LLMs do not directly process from content. A brand consistently mentioned across dozens of publications in plain text is more visible to a language model than a brand with many incoming links but sparse textual mentions.
Ahrefs confirmed this with data at scale: an analysis of 75,000 brands found branded web mentions have a Spearman correlation of 0.664 with AI Overview citation rates. Backlinks correlated at 0.218. The gap is 3x. This is the core divergence between Google SEO and ChatGPT SEO.
What actually predicts ChatGPT citation
The research converges on five variables with consistent positive correlation to ChatGPT citation rates. Each is independently confirmed by multiple studies.
1. Author credentials and attribution
Presence AI tracked 1,200 pages and 3,600 queries across ChatGPT, Perplexity, Google AI Overviews, and Gemini over 90 days. Pages with expert authors and documented credentials achieved a 72% AI citation rate. Pages with no author attribution achieved 25%. Pages with partial attribution (name only, no credentials) achieved 41%.
The practical implementation: every published page needs a visible author byline, a credentials statement ("Senior researcher at X", "10 years in Y"), and a link to an author page. AccuraCast's analysis of 9,000 citation sources found Person schema appeared in 58.9% of cited pages and 70.4% of ChatGPT citations specifically.
2. Content structure: headings and extractability
AirOps analysis of 815,484 retrieved pages found that 7 to 20 subheadings was the optimal range for per-query citation rate in ChatGPT. Pages with fewer than 7 headings (insufficient structure) and pages with more than 20 (over-fragmented) both underperformed.
The mechanism: ChatGPT retrieves approximately 6 to 7 pages for every one it cites. 85% of retrieved content is discarded. Clear heading structure creates extraction points — the model can locate the exact section that answers the query and extract it cleanly. Pages structured as walls of prose with no internal navigation are harder to mine for specific answers.
3. Branded web mentions
Branded web mentions — references to your brand or domain by name across third-party publications, with or without a hyperlink — are the strongest single predictor of AI citation rate across platforms. The Ahrefs 0.664 correlation exceeds backlinks (0.218), domain rating, and organic traffic as predictors.
Building branded mentions requires digital PR: being quoted as an expert source in industry publications, having your research cited by other writers, securing brand mentions in round-up articles. The goal is to appear by name in text that AI systems train on and retrieve from — not just to acquire hyperlinks.
4. Content freshness
ChatGPT with browsing enabled weights recent content more heavily. Amsive analysis found 50% of AI-cited content is under 13 weeks old. BetterAISearch's own freshness decay calibration, which is based on this and corroborating data, gives full weight to content under 3 months old and discounts content over 18 months by 60%.
Practically: every key page on your site should have a visible, accurate last-updated date. Updating with new data or revised findings — even minor additions — resets the freshness signal. Stale content on active topics is a direct citation liability.
5. Specific, sourced claims
ChatGPT is optimised to produce accurate responses. It prefers sources whose claims can be verified against its training corpus. Content with specific numbers, named studies, documented methodology, and properly attributed quotes gives ChatGPT more confidence that it can cite you accurately without hallucinating details.
Generic advice ("content should be high quality") creates no extractable evidence for ChatGPT to work with. Specific claims ("AirOps found 815,484 pages; 500 to 2,000 words outperformed") give the model something concrete to cite and verify.
ChatGPT technical requirements: what crawlers can and cannot read
GPTBot, OpenAI's web crawler, does not execute JavaScript. A Writesonic study of 62 webpage elements across six AI crawlers found that JSON-LD structured data scored zero out of six for readability — none of the crawlers reliably extracted it. JavaScript-rendered content scored 1 out of 6.
The content GPTBot can read: visible body text (6/6), semantic headings H1–H3 (6/6), title tags (5/6), and image alt text (4/6). Everything else is either not reliably read or not read at all.
The checklist for ChatGPT technical eligibility:
- robots.txt must allow GPTBot:
User-agent: GPTBot,Allow: / - Primary content must be in server-rendered HTML, not JavaScript-populated
- Title tag must accurately describe the page content
- H1–H3 heading structure must reflect the page's informational architecture
- No login walls or cookie gates blocking content access
The ChatGPT SEO vs Google SEO action table
| Tactic | Google SEO impact | ChatGPT citation impact | Priority shift |
|---|---|---|---|
| Backlink building | Very high | Low (0.218 correlation) | Downgrade |
| Branded web mentions | Medium | Very high (0.664 correlation) | Upgrade significantly |
| Author credentials | Indirect | Very high (2.4x citation rate) | Upgrade significantly |
| Person schema markup | Low | High (58.9% cited page rate) | Upgrade |
| Content freshness | Moderate | High (50% of cited content <13wk) | Upgrade |
| Keyword density | Moderate | Not a direct signal | Downgrade |
| H1–H3 heading structure | Moderate | High (6/6 crawler readability) | Maintain |
| JSON-LD schema markup | High | Zero (0/6 crawler readability) | Redirect to Google only |
| Word count (long-form) | Moderate | Volume, not rate (use strategically) | Nuance required |
Source: Ahrefs (75k brands), Presence AI (n=1,200 pages), Writesonic (62 elements, 6 crawlers), AirOps (n=815,484 pages)
The bottom line
ChatGPT SEO is not Google SEO with a different name. The source pool barely overlaps. The signals that drive ranking in Google — backlinks, domain rating, keyword density — have weak or zero correlation with ChatGPT citation rates. The signals that drive ChatGPT citation — branded web mentions, author credentials, content freshness, heading structure — are either secondary or absent from traditional SEO practice.
The research is now consistent enough to act on. Start with author attribution (highest measured impact per unit effort), branded web mentions (highest correlation overall), and technical accessibility (binary — either GPTBot can read your pages or it cannot).