The Debrief
L7L14L30L90All
PaidSearchIndustryTechDataBrandConversion
Search · 3 min read23 May 2026

LLM Guidance Does Not Port Across Models the Way SEO Guidance Did. The Optimisation Playbook Just Got Six Times Harder.

Duane Forrester's new Search Engine Journal piece argues optimisation guidance no longer transfers the way SEO guidance did across Google, Bing and Yahoo. Each LLM trains on different corpora, runs different crawlers and applies different alignment. Brands optimising for one model are blind to the rest.

Following Google's guidance about Gemini will leave you optimised for a slice of the LLM landscape and blind to the rest.

3 min read

Duane Forrester published a piece in Search Engine Journal this week that says what every honest SEO has been muttering for six months. LLM optimisation guidance does not transfer between models the way SEO guidance transferred between search engines. The reflex of "build for Google, ship for everyone" no longer works.

For about two decades, SEO worked on a comfortable assumption. If Google said XML sitemaps mattered, Bing said XML sitemaps mattered. If Google rewarded structured data, Bing rewarded structured data. The discipline was effectively single-engine because the other engines copied. Practitioners could optimise for Google with reasonable confidence that the work carried.

That world is gone. The major LLM providers train on different corpora. They run different crawlers under different policies. They route different queries through different retrieval systems. They apply different alignment processes. The optimisation surface is six platforms (ChatGPT, Gemini, Claude, Perplexity, Copilot, Grok), not one. Each one has its own discovery, citation and ranking model.

That is not a failure of any single provider. It is the structural difference between the LLM era and the search era.

Why it matters

Australian brands building AI search strategies are mostly doing one of two things. They are extrapolating from Google's guidance (which only covers Gemini and Google AI Overviews). Or they are deferring to a single visibility tool that reports a generalised "AI mentions" number that does not distinguish between models. Neither approach holds up.

The practical reality is that ChatGPT may cite your brand consistently while Gemini ignores you, or Perplexity may show your domain in the source list while Claude pulls from competitors. The visibility profile per model is different. The optimisation per model needs to be different too.

6 models

The minimum number of LLM surfaces marketers now need to track separately for citation, source attribution and answer inclusion

What to do about it

Set up monitoring per model, not in aggregate. Test the same five buying-intent queries in ChatGPT, Gemini, Claude, Perplexity, Copilot and Grok. Capture the answer and the cited sources. Do it monthly.
Identify which models cite your brand and which do not. The pattern usually reveals which corpora the model favours (Reddit-heavy, Wikipedia-heavy, news-heavy, primary research-heavy).
Build for the corpora that matter to your category. If ChatGPT relies on Reddit and your competitor has 12 subreddit mentions to your zero, that is the work. Not another schema tweak.
Differentiate the discovery work from the citation work. Getting indexed is one job. Getting cited is another. Different models reward different things at each stage.
Stop expecting one tool, one report or one agency framework to give you a unified AI search strategy. The unified product does not exist yet because the platforms have not converged. Treat each LLM as a separate distribution channel.

The broader implication is that the era of single-engine SEO is over and the brands that figure out the per-model playbook first will set the standards the rest of the market eventually copies. The window is roughly 18 months. After that the major LLMs will start to converge on similar discovery rules and the advantage will compress.

Share this brief
Send it to a colleague who'll find it useful.
Filip Ivanković
The Debrief / From Filip Ivanković
One every morning. Six months in, you'll see the patterns most don't.
Strategy, benchmarks, and what's actually moving in Australian marketing. Four-minute read. The reps compound.
Filip Ivanković·Founder, New RebellionLinkedIn