Local SEO blog
Someone Rated 193,000 AI Answers. The Pages That Get Cited Start With the Answer.
Last updated
A study of 193,507 AI answers found a simple pattern: pages cited in AI answers started with a direct answer more often than comparable pages that went uncited. In one blind comparison, cited pages led with an answer in 23% of pairs, compared with 14% for uncited pages surfaced in the same AI answer. Against Google top-10 pages the AI never cited, the split was 20% to 5%, according to TJ Robertson’s study and Search Engine Journal’s coverage.
I’ve been arguing for answer-first content for years. This study reads like a useful confirmation of what I’ve been building, with an important caveat: it shows an association. It does not prove that opening with a direct answer causes an AI system to cite a page.
What the study looked at
Robertson published the study in late October 2026. It analyzed 193,507 answers from ChatGPT, Google AI Mode, and Google AI Overviews. The answers were collected from August 27 through October 21, 2026, using buyer questions tracked by his agency for its clients.
The 23% versus 14% comparison looked at pages surfaced in the same AI answers, with cited pages compared against pages that appeared but were not cited. In a separate blind-rating round, Robertson compared cited pages with pages that ranked in Google’s top 10 for a related question but never appeared among the AI’s sources. In that comparison, 20% of cited pages began with a direct answer, compared with 5% of the passed-over pages.
The blind ratings matter because the raters assessed the pages without knowing which ones had been cited. They looked at how the pages opened and how well they answered the question.
The limits matter just as much as the headline. The study used questions tracked by one agency across its client base. That gives us a substantial dataset, but it does not represent every industry, question, website, or AI search experience.
Most importantly, the study compares page traits. It does not show what happens when someone edits a page to put the answer first and then measures citation changes against a control group. Robertson makes that distinction himself. The finding is an association, not a causal result.
That kind of honesty is useful. A big sample can help identify a pattern worth testing. It cannot turn an observational study into a controlled experiment.
Why answer-first content makes sense
An AI-generated answer needs information it can retrieve, interpret, and use. A clear opening gives a system a compact passage that directly addresses the question. The rest of the page can add context, examples, qualifications, and helpful next steps.
I’ve written before about keeping answer-ready passages concise, including the roughly 200-character framing in my post on what makes ChatGPT cite pages. The point is straightforward: make the useful answer easy to find and understand.
That matches my content principle: answer the question before asking for the sale. A service page can explain the service right away, then tell the reader how to take the next step. An educational page can provide the explanation first, then connect it to a product or service where it makes sense.
This is also part of the larger work of organic growth. Useful pages can bring in search visibility and help readers understand a topic before they’re ready to act. Clear structure and sound crawlability support that work, too. My technical SEO and AI search optimization pages cover those parts of the foundation.
I care more these days about showing up inside LLM answers than staring at another rank report. Rankings still matter, and I still use tools such as Google Search Console, Google Analytics, Ahrefs, and Semrush. Citation visibility gives me another way to see that useful information is reaching people where they’re asking questions.
Give your five most important pages a five-minute audit
You don’t need a tool for the first pass. Open the five pages that matter most to your business and read the first lines as if you were a potential customer.
- Name the question the page is meant to answer. A product page might explain what the product does. A service page might explain what the service includes.
- Read the opening paragraph by itself. Can someone understand the direct answer without scrolling or reading the introduction twice?
- Find any throat-clearing. Long setup, brand history, and broad scene-setting can push the useful answer too far down.
- Move the answer up. Put a direct, accurate response near the beginning, then use the rest of the page to explain details and context.
- Keep the answer complete. Add important limits or qualifications right there when leaving them out could mislead someone.
A quick example: if the page answers “What does a technical SEO audit cover?”, the opening should explain what it reviews. The supporting sections can then cover crawlability, indexing, site structure, and other details. A reader should get the basic answer before the page starts asking for a call.
Keep the wording natural. A page written for real people tends to be clearer than one stuffed with the same search phrase over and over. The goal is a useful opening that fits the question and gives readers enough context to trust what follows.
If you find several pages burying the answer, start with the one closest to a buyer’s decision. Fix that one, then review the others. If you want help finding content and technical priorities, my free SEO audit is a practical place to start.
A small test with a clear signal
I’ve worked in SEO since 2013, and the answer-first approach has been part of how I build content for years. Robertson’s results give that approach a timely signal across a large collection of tracked AI answers. The numbers are worth paying attention to, and the study’s limits belong in the same conversation.
Start with your five pages. Read the first paragraph. Check that it answers the question before the page asks the reader to buy, book, or contact you.
Answer-first is the cheapest test on the board. It takes a few minutes, requires no special tools, and happens to be the approach this study supports.