ChatGPT draws its citations from a completely different pool than Google. Averi.ai measured it across 680 million citations: the SERP overlap is 4.2%. That number went down when GPT-5.5 launched in April 2026. GPT-5.6 went live August 6.

If you're running one unified SEO strategy and expecting it to handle both Google and ChatGPT, you're covering 4.2% of the target with the wrong tool.

What the Data Shows

Averi.ai's May 2026 study is the largest citation analysis published. 680 million citations, five major AI platforms, cross-referenced against Google SERP data. The platform-by-platform SERP overlap numbers:

Gemini: 41.1% — heavily SERP-aligned, as expected from a Google product
Perplexity: 39.4% — also SERP-adjacent, favours real-time indexed sources
Microsoft Copilot: 23.7% — Bing-derived, moderate correlation
Grok: 14.1% — social-first, low SERP correlation
ChatGPT: 4.2% — effectively independent of Google

The September 2025 citation shift confirmed this structurally. Google removed the num=100search parameter from its API. Two foundational ChatGPT citation sources collapsed 80% within six weeks. The mechanism: ChatGPT's web retrieval pipeline is tightly coupled to Google's organic rankings when it does search — but ChatGPT only triggers web search when it decides the query requires it (typically 3-6 links per response). The result is a pipeline that looks like Google when it runs, but runs far less often than practitioners assume.

The Everything PR ChatGPT Citation Index 2026 adds further context. Reddit gained significant citation share after GPT-5.5 — counter to what most GEO strategies optimise for. Community-sourced content, forum threads, and UGC now outperform many brand domains on ChatGPT.

Why This Matters

The implications are binary. Either you treat ChatGPT as a separate channel requiring its own strategy, or you accept that your current SEO coverage leaves 95.8% of ChatGPT's citation pool untouched.

Domain authority proxies are weak ChatGPT predictors.Traditional SEO relies on backlink profiles, domain rating, and topical authority — signals Google weighs heavily. ChatGPT's 4.2% SERP correlation means these signals transfer at roughly 1/24th the rate they do on Google. A DA-80 site is not automatically a ChatGPT citation candidate.
Model updates reshape citation sources every 2-4 months. The GPT-5.5 update in April 2026 redistributed 47% of citations. GPT-5.6 launched August 6. Any 12-month ChatGPT visibility strategy built on current source rankings is modelling stability that does not exist. The correct planning horizon for ChatGPT citation strategy is one model cycle, not one year.
Reddit and community signals beat brand signals. Post-GPT-5.5, Reddit gained the most citation share — a counterintuitive result for brands investing in owned-channel content. ChatGPT pulls heavily from sources where real users answer real questions in natural language, not brand copy or press releases.
Only 11% of ChatGPT and Perplexity citations overlap. The same Averi.ai dataset shows just 11% shared domains between ChatGPT and Perplexity. One GEO strategy cannot serve both platforms simultaneously — the citation pools are almost completely separate.

What to Do

Audit your query-type mix.ChatGPT only triggers web search when it decides the query requires it. Informational queries, comparison queries, and “what is X” questions are more likely to pull from training data than current citations. Identify which queries in your category trigger web retrieval — those are the ones where fresh, crawlable content matters. Queries answered from training data require a fundamentally different approach: entity establishment, not link building.
Build third-party mentions across Reddit and community platforms.Not forum spam — genuine participation, sourced expertise, and content that gets organically cited in threads. The goal is to appear in places ChatGPT treats as high-signal. Reddit is one; Quora, Stack Overflow, and niche community forums vary by category. Identify where your category's authentic Q&A happens.
- Prioritise threads where your category gets compared — those generate multi-source citations
- Answers from named experts with public profiles outperform anonymous posts
- Link to your own content from community posts sparingly — value-first, not link-first
Decouple your ChatGPT KPIs from your Google KPIs. Ranking position on Google does not predict ChatGPT citation frequency. Set separate measurement frameworks:
- Google: track ranking, CTR, organic sessions
- ChatGPT: track citation frequency by query category, source attribution, and share of voice in AI responses (use tools like Profound, Scrunch, or manual prompt sampling)
Plan around model update cycles. GPT-5.6 is live now. The next major update is likely within 2-4 months. Run a mini citation audit immediately after any GPT release — track which sources gained and lost share. This is the ChatGPT equivalent of tracking a Google core update. Set a calendar reminder for October 2026.
Optimise for query types that trigger web search.ChatGPT's web retrieval activates for: recent events, pricing data, product comparisons, and local queries. These are your highest-value citation opportunities — fresh, crawlable content that answers these query types directly is your ChatGPT content priority, not long-form brand narratives.

What We Changed

"Run a separate strategy" is easy to write and expensive to actually do, so here is what separating it cost us — and the rule we now enforce on ourselves as a result.

// one rule set, run against us before it runs against you

The checks Chad reports to a customer and the checks that gate our own deploys are the same code, not two lists that agree in principle. We had already learned what happens otherwise: our internal gate checked only whether titles were too long, so it reported 28 over-length titles on this site and stayed silent about 12 that were too short — the smaller number, and the more damaging fault.

Two rule sets disagreeing about the same site is worse than one rule set with gaps, because both look authoritative. So there is now one copy, and every rule we report to a customer has already been run against trychad.ai.

The separate-strategy finding also produced a rule we did not enjoy adding. If ChatGPT and Google are citing different pools, the tiebreaker is whether your page contains something the original sources do not — and a research write-up, by construction, does not.

// content/information-gain — added, then immediately failed by us

The rule flags any article whose every source belongs to somebody else. It ran against this site and reported four pages — which was every page on the site that cited anything at all.

One of the four was our own post arguing that information gain is what decides whether you get cited. A daily research-to-blog pipeline produces this by default: it reads other people's studies, so other people's studies are what it has to cite. Nothing was done wrong — the loop simply had no step that asked and what do we know?

This post is the first output of that loop having one. Whether it is worth citing over Averi's original 680M-citation analysis is a fair question — and the honest answer is that it is only worth it for the part above, which is ours.

Chad's Take

The uncomfortable truth: most companies are not running a ChatGPT strategy at all. They're running a Google strategy and assuming it covers ChatGPT by proximity. It covers 4.2% of it. The 95.8% gap is not a growth opportunity — it's a current liability. Every competitor who builds a genuine ChatGPT citation presence while you wait for your SEO rankings to transfer is stacking share in a channel where your traditional authority means almost nothing. The question is not whether to build a separate ChatGPT strategy. It is whether you do it before the next model update reshapes the field again.

Sources

TryChad.ai — content/information-gain: a build rule flagging articles whose every source is third-party. First run reported four pages — every page on this site that cited anything at all, including the post arguing information gain decides citation, August 2026. Primary source.
TryChad.ai — Single-source rule set: the checks reported to customers are the same code gating this site's deploys, adopted after two divergent copies disagreed about the same site, August 2026. Primary source. You can run them against your own domain.
Averi.ai — AI Citation Platform Divergence Analysis across 680 million citations from ChatGPT, Gemini, Perplexity, Grok, and Microsoft Copilot, May 2026. averi.ai
Everything PR — ChatGPT Citation Source Index 2026, tracking citation source redistribution across GPT-4o, GPT-5.5, and GPT-5.6 model cycles, August 2026. everything-pr.com
GitHub / danishashko — AI Citation Patterns Study, open dataset tracking cited sentence characteristics and positional bias across major AI platforms, July 2026. github.com/danishashko/ai-citation-patterns
OpenAI — ChatGPT Release Notes, GPT-5.6 launch August 6, 2026. help.openai.com

// is your content in the 4.2%, or only in Google's half?

Chad scores ChatGPT, Perplexity, Gemini and Google AI Overviews separately — because after this finding, one number across all four is an average of two different games. Same rules we run against our own build, including the information-gain check above. Run it on your domain.

Ask Chad how you look →