Search Engine Journal covered research by TJ Robertson of TJ Digital. The sample included 193,507 responses from ChatGPT and Google AI systems (AI Mode and AI Overviews). Pages cited by AI models did contain direct answers in opening lines more often.
But the gap between cited pages and those found but not used by AI was small. Robertson does not call direct answers a cause of citation — more a correlation.
Site-wide rewrites are unnecessary
The takeaway is simple: reworking all content is wasted effort. Start by checking priority pages. Each one against its target question.
The logic: if a page aims for an AI answer slot, it should deliver upfront. But that does not mean other elements can be ignored.
How to check page openings
The process relies on logs of AI model response measurements and citation audits. This is operational work that fits into SEO processes.
What to do: compile a list of priority pages — those already driving traffic or addressing common questions. For each, formulate the user's target question. Check whether the first paragraph gives a direct answer. Record whether AI systems currently cite the page.
Why this works selectively
AI models weigh more than content structure. Source authority, full-page relevance and markup quality all matter. A direct answer at the top is one signal, not the only one.
If a page is weak on other parameters, rewriting the opening will not pull it into citations. For strong pages, it may act as a trigger.
Where to start
Check a few pages manually. Ask their target questions to ChatGPT or Google AI. See if the system cites your page. If not, analyse the opening: is there an answer, is it clear without context.
Then scale. Citation audits can be automated through model APIs or monitoring services. Track the pattern: which pages get into answers, which drop out.
Practical approach
Optimisation for AI is not a separate strategy. It is part of content work. The page opening should answer the user's question whether a human or a model reads it.
Checking priority pages is a logical first step. Mass site rework without measurement and analysis is not. Robertson's data show correlation, not causation. Act selectively and measure results.
