AI assistants and Google AI Overviews cite a handful of pages per answer. Getting into that handful is a page problem first and a site problem second. A technical check sits behind both. Most businesses only have the first one wrong.
How answer engines pick sources
An AI Overview is built from pages Google already ranks for the question and the related questions around it. Google explains this in its AI features documentation. ChatGPT and Perplexity run a search, read the results and cite the passages that answer the question most directly.
In every case the machine is looking for a passage it can quote or paraphrase with confidence. A plain statement of fact, with a source, under a heading that matches the question. Pages that make the reader work for the answer do not get cited.
The page
Put the question in the heading, in the words a customer uses. Answer it in the first two sentences. Forty to sixty words is the length that gets lifted whole.
Follow with the detail, one idea per paragraph. Tie every number to a source a reader can check. Define each technical term the first time it appears. Add a short set of related questions at the bottom, each answered in the same direct way.
Behind the page, add Article structured data and FAQPage structured data for the questions. That tells the machine what it is reading without guessing.
The site
One page rarely gets cited on its own. Answer engines favour sites that cover a topic in depth, because depth is evidence the site knows the subject. Build a parent page for the topic and a cluster of question pages underneath it. Link each question page up to the parent and sideways to its neighbours.
Keep facts that change in one place. On Gridly, rebate amounts and electricity prices live in a single data file, and every page reads from it. A stale number on one page costs trust across the site.
The technical checks
Open your robots.txt file. Many sites block AI crawlers by accident, through an old blanket rule or a plugin default. OpenAI documents its crawler in the GPTBot guide. Google-Extended controls Gemini’s use of your pages. Perplexity publishes its crawler names too. Allow the ones you want cited.
Then check that the pages render without JavaScript, load in under about two seconds and carry a canonical tag. AI crawlers are less patient than Google’s.
Off the page
AI assistants weight how often a business is named across the web. Keep your name, category and description consistent in every directory and profile. A business described five different ways reads as five weak signals instead of one strong one.
How to measure it
Google Search Console shows impressions and clicks, including from AI Overviews, but does not split them out. Ahrefs and similar tools now count the AI answers that cite each page, by platform. On Gridly in September 2026 that was about 1,900 answers. 258 came from Google AI Overviews, and the rest were split across Copilot, AI Mode, Perplexity, Gemini and ChatGPT.
Watch both numbers. A page can be cited often and clicked rarely, because the answer engine gave the reader what they needed. That is still a page doing its job, and the citation puts your name in front of the buyer.
What not to do
Do not publish hundreds of pages with the place name swapped. Do not stuff the phrase you want to rank for into every paragraph. Do not hide the answer behind an introduction that restates the question. Each of those reads as low value to a machine built to find the best answer.
If you want this built rather than explained, the method and the numbers are on our AI SEO agency page.