SEO teams lose hours to sorting: query exports, page lists, decay checks, briefs. AEO adds a new question on top. AEO means getting named when someone asks an AI assistant. If a buyer asks ChatGPT or Google’s AI Mode about your category, what does it say, and does it mention you?
The reassuring part is that Google doesn’t ask for much that’s new. Its documentation says: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” A page has to be indexed and eligible to show with a snippet. That is ordinary SEO. So the first gain from AI is time back on the old work.
I did this work at Gaba Healthcare. Organic visitors grew from 5,000 to 9,000 a month, with patient-education articles grouped by topic, and SEO brought in 35% of qualified calls. Several changes ran at the same time, so I can’t split the credit. Sorting and checking is the part I’d now give to AI. I haven’t run an AEO programme, so the AEO workflows below are plans.
Google’s documentation lists things you don’t need for its AI features:
llms.txt or other AI text files. “You don’t need to create new machine readable files, AI text files, or markup.”
Special schema. There is “no special schema.org structured data that you need to add.”
Rewriting or chopping pages into small pieces. Coverage of Google’s May 2026 guide says its systems can find the relevant part of a page. That’s a secondary report, so check Google’s own page.
This site has an llms.txt file. It costs nothing, but Google says you don’t need it, and I haven’t found independent evidence that it changes what other assistants cite. I wouldn’t spend a day on it.
1. Group search queries by intent
In: twelve months of Search Console queries, with pages, clicks and impressions.
AI does: groups queries by what the searcher wants (learn, compare, buy), flags pairs of your pages that compete for one query, and lists query groups with impressions but no page.
A person checks: someone who knows the product reads every group, merges the bad ones, and rejects any “gap” you already cover under another name.
Measure: hours to produce the content plan, and the share of planned pages that earn impressions within 90 days.
2. A refresh queue for pages that are slipping
In: page-level clicks by month, and the current text of each page.
AI does: ranks pages whose clicks have fallen over three months and, for each, lists what is out of date or thin: old figures, a missing question, dead links. It doesn’t rewrite the page.
A person checks: a writer adds something that only you have: a number from your own data, a real customer question, a screenshot. Coverage of Google’s guide says it asks for “non-commodity content”, meaning more than anyone could write from common knowledge.
Measure: clicks to refreshed pages in the eight weeks after, against the eight weeks before, with a few similar untouched pages as a comparison.
3. A weekly check of AI answers
In: 20 to 30 real questions buyers ask, taken from sales calls, support tickets and Search Console.
AI does: asks each question in the assistants you care about, and logs whether you’re named, which sources are cited and how the answer describes you. Check each tool’s terms before automating this. Where they don’t allow it, do it by hand. It’s a short list.
A person checks: someone reads the log for wrong or stale statements about you, and picks the three that matter most.
Measure: the share of questions where you’re named, and the wrong statements fixed. Answers change from run to run, so I’d read a month of logs as a trend and not trust one run.
I haven’t found an independent study that shows which page features cause a citation. The large citation studies come from companies that sell tracking tools, and their findings suggest that different assistants cite different sites. I use them for ideas and check each assistant myself.
4. A crawl-access audit
In: your robots.txt, the pages that matter most, and the Search Console indexing report.
AI does: reads robots.txt and robots meta tags, and lists which crawlers can reach which pages, plus pages that aren’t indexed or carry a nosnippet rule.
A person checks: each block is a decision someone made on purpose. Two facts to know. For Google’s AI features, a page must be indexed and snippet-eligible. For ChatGPT, OpenAI says OAI-SearchBot is the crawler that surfaces sites in its search results. A blocked site “will not be shown in ChatGPT search answers, though can still appear as navigational links.” GPTBot is a separate crawler for training, and blocking it doesn’t affect search.
Measure: important pages that are blocked or unindexed by accident, which should be zero.
5. Find where the answers get their facts
In: the log from workflow 3, with the URLs each answer cited.
AI does: groups the cited sources by type (your site, review sites, forums, news, competitors), and lists the questions where you’re missing and who is cited instead.
A person checks: whether each gap is real, then picks the honest response: improve your own page, publish your data, correct an outdated listing, or ask customers who really use the product to review it. Never write or buy reviews. The FTC’s final rule bans fake reviews, including AI-generated ones by “someone who does not exist”.
Measure: gaps closed, and questions where you move from missing to named over three months.
Hours saved on the content plan, clicks to refreshed pages, the share of buyer questions that name you, and wrong statements fixed. For tools: Google says its Search Generative AI performance reports show impressions in AI Overviews, AI Mode and Discover by page, country, device and date. Bing’s AI Performance report, in public preview since February 2026, shows citations by page and query. Neither shows ChatGPT, so for that I’d use your analytics referral report and the weekly log.