Adapted from Duane Forrester's analysis on Duane Forrester Decodes. Aurora extends his framing for marketing teams running AI-search visibility programs.
For twenty years, SEO had a load-bearing assumption underneath it: what you did for Google would mostly work for Bing. The reason was not luck. It was infrastructure. Google, Yahoo, and Microsoft formally agreed to support Sitemaps 0.90 in November 2006. Five years later they launched Schema.org together (Yandex joined later). robots.txt was eventually formalized as RFC 9309 in 2022. IndexNow was launched by Bing and Yandex in 2021 and picked up by Naver, Seznam, and Yep. The engines were competitors. They built shared standards anyway, because shared standards served them all.
That overlap layer is the reason a practitioner could follow Google's guidance with confidence that the underlying signals would carry to other engines. The LLM ecosystem does not have that layer, and the absence is structural, not temporary.
Where the LLM stack actually diverges
Training data. OpenAI has disclosed licensing deals — News Corp (up to $250M over five years), Axel Springer ($13M yearly), Reddit ($70M annually), plus the FT, AP, Le Monde, and others. Google has its own Reddit deal valued around $60M annually. Anthropic has not disclosed equivalent arrangements. Nobody outside these companies knows what any provider has paid for and what it has not. The training corpus is provider-specific and partially private.
Crawler infrastructure. OpenAI runs GPTBot (training), OAI-SearchBot (indexing), and ChatGPT-User (retrieval). Anthropic runs ClaudeBot, Claude-SearchBot, and Claude-User. Perplexity runs PerplexityBot and Perplexity-User. Google introduced Google-Extended in September 2023 to control Gemini training separately from Googlebot. There is no single AI user-agent. Every provider needs a separate robots.txt entry. Block one and you do not block the others.
Retrieval architecture. ChatGPT has historically used Bing as its primary web index. Perplexity built on a Vespa pipeline that treats document chunks as first-class units. Gemini uses Google's own index plus Knowledge Graph grounding. Claude partners with Brave Search. The same query, four different result sets.
Alignment methodology. Post-training shapes behavior in ways that show up in actual outputs. OpenAI leans on RLHF. Anthropic developed Constitutional AI, training models to critique and revise against written principles. Identical retrieved content fed to two differently-aligned models can yield materially different responses about the same brand.
llms.txt is the case study
Jeremy Howard proposed llms.txt in September 2024 as a markdown manifest at the site root. The SEO community embraced it. By mid-2026, no major provider had confirmed consuming the file. Server-log analyses across hundreds of thousands of domains showed major AI crawlers do not routinely request /llms.txt. Google's John Mueller compared it to the deprecated meta keywords tag. Gary Illyes confirmed at Search Central Live in July 2025 that Google does not support and does not plan to support llms.txt. Schema.org succeeded because three engines built it together and enforced it together. llms.txt was proposed by one researcher, picked up by tooling vendors, and ignored by the platforms it was supposed to serve.
The Gemini inversion
Google publishes SEO documentation emphasizing classical ranking signals — E-E-A-T, content quality, technical accessibility. That guidance works for Google Search itself. It does not necessarily work for Google's own AI surfaces. In late 2024, roughly three-quarters of AI Overview citations also ranked in Google's top 12 for the same query. By early 2026, after Gemini 3 shipped in January, Ahrefs analyzed 4 million AI Overview URLs and found only 38% of cited pages appeared in the top 10. BrightEdge measured the overlap closer to 17%. SE Ranking saw Gemini 3 replace 42% of previously cited domains and generate 32% more sources per response.
AI Mode widens the gap further. SEMRush data shows AI Mode and AI Overviews reach semantically similar conclusions 86% of the time but cite the same URLs only 13.7% of the time. Only 14% of AI Mode citations rank in Google's traditional top 10. The same brand, the same content, three different outcomes across Google Search, AI Overviews, and AI Mode.
Qwairy's analysis of 118,000 responses across ChatGPT, Perplexity, Google AI Mode, and Claude is the punchline: only 11% of cited domains appeared across multiple platforms. The other 89% were platform-specific.
What this means for the work
The old reflex was to optimize for Google and trust the portability. The new discipline is to test visibility per platform and treat divergence as the default. Aurora is built around that discipline:
- Every prompt runs against every engine you select. ChatGPT, Perplexity, Gemini, Google AI Mode, Google AI Overviews, Copilot, Claude — independently, on a schedule.
- Per-engine citation and visibility breakdowns make platform-specific gaps visible. A brand strong on Perplexity and invisible on Claude shows up as exactly that.
- Source intelligence per engine surfaces which authority domains each platform reaches for in your category. The overlap between them is usually smaller than teams assume.
- Adaptive querying catches the moments a model upgrade shifts the geometry. Gemini 3 was a 42% citation churn event. You want to see it the week it happens, not the quarter after.
The overlap layer shrunk. You now have more work to do, not less. The practitioners who recognize that first will spend the next two years setting the standards everyone else follows.
