Most advice about getting cited by AI search engines is theory written by people who have never measured it. So we ran the experiment on ourselves. Every week we test the same 10 commercial-intent queries across AI search surfaces, record exactly which domains get cited, and diff the results against the prior week. This post is the first three weeks of that dataset, including the part where we went from invisible to page one on a commercial query in seven days, and the nine queries where we are still nowhere.
What was the experiment?
We picked 10 queries a real buyer would ask, things like “best AI SEO agency for law firms,” “AEO services pricing,” and “how to get cited by ChatGPT as a law firm.” Once a week we run all 10, log which domains appear in the AI synthesis, which pages earn citations, and what changed week over week. Week one we appeared on 1 of 10 queries (our own brand query). Week three we appear on 2 of 10. That sounds small because it is. AI visibility from a cold start is slow, and anyone telling you otherwise is selling something.
What actually moved the needle?
One blog post. In week two we published a guide structured the way AI engines extract content: every H2 is a literal question, the first 40 words under each H2 answer that question directly, supporting data sits in HTML tables, and the page carries proper schema markup. Seven days later that post sat at roughly position 4, page one, inside an answer cluster alongside agencies that have been working these queries for years.
Nothing else about our site changed that week. Same domain authority, same backlinks, same everything. The page structure did the work.
What does the citation data say across all 10 queries?
Three patterns held across every week of tracking so far:
First, the engines cite pages, not brands. The domains winning our tracked queries (Juris Digital, First Page Sage, ALM Corp, Marcel Digital) win with dedicated, narrowly scoped pages, one per query intent, not with their homepages. The clearest example: one competitor now tops two separate AI-search queries with two separate dedicated service pages, one per AI surface.
Second, you do not need to rank top 10 to get cited. One stat that surfaced repeatedly in tracked syntheses: 62% of links cited in Google AI Overviews come from outside the traditional top 10 organic results. That is the entire argument for answer-formatted content. The extraction layer rewards different things than the ranking layer.
Third, the syntheses lean on extractable numbers. Pages that state specific figures in plain sentences (“firms appearing in 70%+ of ChatGPT responses average 15+ entity citations and 50+ reviews at 4.7 or higher”) get quoted. Pages that say “we drive results” do not exist as far as AI engines are concerned.
How long does it take to get cited from a cold start?
Based on our own data and what we see running this for clients: the first movement on a single well-built page can happen in one to two weeks (Perplexity and Google AI Mode refresh fastest), but meaningful coverage, appearing on several commercial queries, takes 60 to 120 days of consistent publishing. Our own scoreboard after three weeks is 2 of 10. We expect to report that number weekly, including the weeks it goes backward.
What page structure should you copy?
The exact template that produced our page-one result:
| Element | What we did |
|---|---|
| H2 headings | Literal questions, phrased the way buyers ask |
| First 40 words per section | Direct, self-contained answer, no throat clearing |
| Data | Specific numbers in sentences and HTML tables |
| Schema | FAQPage plus Organization, service schema where relevant |
| Scope | One query intent per page, not one page for everything |
That is the whole trick. It is not secret, it is just labor.
What didn’t work?
Worth being honest about the misses. Eight of our ten queries still show zero SubscribePR presence after three weeks, including the highest-value ones (“best AI SEO agency for law firms”). The incumbents on those queries hold positions built on years of authority, reviews, and directory presence, and one structured page does not displace that. Those queries need the slow stack: consistent content, press coverage, directory consistency, and review volume. We will keep publishing the weekly numbers either way.
If you want to run this same tracking on your own business, the method is free: pick 10 queries your buyers ask, run them weekly in the AI engines, log the cited domains in a spreadsheet, and diff. The data tells you exactly which page to build next.
Tagged