Generative Search Optimization
There is a lot of confident advice about getting cited in AI answers and very little sorting of it. This is our attempt at that sorting: what has evidence, what is a reasonable bet, and what has no mechanism behind it at all.
What changed, precisely
Ranking stopped being sufficient. Ahrefs put the overlap between Google's top ten and the pages AI Overviews cite at 76 percent in July 2025, and measured it at 38 percent since. At the same time answers began drawing more heavily on video, community and reference platforms, which matches how Google describes the sourcing behind its AI features in Search.
Ranking is still a strong input. It is no longer the gate. A company can hold position one for its category term and be absent from the answer displayed above it.
What has the strongest evidence
These are the ones we would act on without hesitation, because they are either mechanically obvious or replicated across independent testing.
| Tactic | Why it holds up |
|---|---|
| Content readable without JavaScript | Most AI crawlers do not execute it and run short timeouts. If the text is not in the raw HTML, there is nothing to cite |
| Answer-first passages | Engines score passage-sized chunks independently. An answer in the sixth paragraph loses to the same answer stated first |
| Cited sources, named statistics, attributed quotes | The largest measured lifts in published testing, and largest for pages not already ranking |
| Consistent entity information | Engines reconcile you from many sources. Contradictory details make that reconciliation fail |
| Genuinely new information | Restating what already ranks gives an engine no reason to prefer you |
| Presence where answers are drawn from | Community, video and review platforms are cited heavily. Absence there is invisible |
Worth doing but unproven
Structured data. Worth having, error-free, for entity clarity and search features. The evidence that it drives citation is weak, and treating it as the lever is a common misallocation.
Freshness. Real but overstated. The "six month cliff" is not supported; cited content skews fresher on average and is frequently a couple of years old. Recency matters most on news-adjacent queries and least on evergreen reference material.
What has no mechanism at all
Prompting engines repeatedly to teach them your brand. This is sold, and it does nothing. Models do not learn from the prompts users type, web search queries are not a training signal, and sessions do not persist. There is no pathway from someone asking ChatGPT about you a hundred times to ChatGPT recommending you later.
llms.txt. No major platform uses it. Monitoring shows essentially no crawler requests for the file. It costs nothing to add and should not appear on anyone's priority list.
Keyword density. It did nothing for traditional search and does nothing here. Testing found no benefit and some risk.
What we do differently as a result
We check Bing. It is the route into ChatGPT and it is the index nobody looks at, so companies with healthy Google coverage are regularly absent from a major answer surface without knowing.
We also measure on a schedule rather than once, because the answers move. A single reading tells you almost nothing; the same questions asked monthly tell you whether anything you did worked.
And we pay most attention to the competitor half of that measurement. Knowing a competitor was cited is trivia. Knowing which specific page was cited converts a vague ambition into a list of things to go and earn.
What to expect
Entity and technical corrections often show within weeks, because you are fixing information rather than building authority. Mention rate commonly moves in 30 to 60 days. Competitive rankings and the organic traffic behind them still take 6 to 12 months. Those three windows are what we observe across our own engagements, not a published benchmark.
Anyone guaranteeing you will be named in an AI answer is describing something they do not control. Engines change outputs continuously, and the honest promise is measurement plus fixing what demonstrably blocks you.
The short version
Make the page readable without JavaScript. Put the answer first. Be specific, cite sources and say something nobody else has. Keep your facts identical everywhere. Be present where answers are drawn from. Everything else is detail, and a few popular tactics are not detail but noise.
The mechanism we use is described under Generative Search Signals, and it sits inside the SEO programme rather than being sold separately.
See where you stand
We will review your current search visibility and show you what buyers find when they look for what you do.
Contact Us