Two New Experiments Show What Gets You Cited by AI

You already know the drill by now, if you've been paying attention to how AI search works: write a Quick read blockquote, use ### Question? headings in your FAQ, keep the page fast, keep it linked to the rest of your site. Do all of that, and you're still watching a competitor's page get quoted by Perplexity while yours — same topic, same effort, similar length — gets skipped.

A new pair of field experiments narrows down one reason why, and it's smaller and more specific than most of the advice floating around: how you phrase the headings that aren't in your FAQ section.

What the two experiments tested

Most GEO advice in circulation right now comes from a single campaign, a screenshot, or a theory that sounds right. Search Engine Land ran something closer to an actual test: two structured experiments, back to back, tracked manually, with the second one deliberately checking whether the first one's conclusions held up somewhere else.

The first experiment ran for an existing brand over several months, tracking 15 commercial-intent keywords across four AI platforms. The second was a 30-day cold start for a small SaaS link-building agency that had no measurable AI presence when the test began, tracking the same 15-keyword structure across six platforms this time. Between the two, the queries generated 775 logged citation events — a real sample, not a handful of anecdotal screenshots.

Two findings held across both tests. Headings phrased as questions — "How is AI search different from traditional search?" instead of "AI Search vs. Traditional Search" — earned more citations than their statement-phrased equivalents. And pages that separated two related-but-distinct topics into their own sections, rather than folding both under one heading, did better than pages that combined them.

A third finding also held: recent data, current references, and a visible publication date correlated with stronger citation performance — the same freshness signal covered here before, now showing up again in an independent, differently-structured test.

The part worth sitting with, though, is what didn't survive: one conclusion from the first experiment failed to replicate in the second. Search Engine Land didn't treat that as a failure of the method — it treated it as the reason the method exists. A tactic that works once, on one site, in one campaign, is a data point. A tactic that survives a second, differently-configured test on a different kind of business is closer to a rule.

Why this matters for a site that isn't a brand's whole marketing budget

Neither experiment was run by or for a small local business, and that's exactly why the finding is useful to one. Nobody is going to burn a few thousand dollars over several months testing GEO tactics on a plumbing company's website. The value of someone else running that test is that the results — question-phrased headings beating statements, freshness beating stale — are cheap to apply and don't require guessing.

This is also a good gut check on the structural skeleton this site already runs on: descriptive H2s, an FAQ in question format, a visible Quick read. That skeleton was never wrong, but it was built around the FAQ section specifically. This is the first solid evidence that the same logic — phrase it the way someone would ask it out loud — pays off outside the FAQ block too, on the ordinary body headings most posts never think to touch.

The mechanics: what to check on your own pages

• Read your H2 and H3 headings out loud as if a customer were asking them. "Pricing" is a label. "How much does this cost?" is a question a person — or an AI system fielding one — would type. Where a heading can become the natural question a reader has, make it one.
• Split combined topics instead of stacking them under one heading. A section titled "Hours and Parking" covering two unrelated things is harder to cite cleanly than two headings, one for each. If a heading has "and" doing a lot of work, that's usually the tell.
• Keep a visible, accurate publication or last-updated date on pages that matter. Not just in metadata — visible on the page itself, the way AirOps' 2026 State of AI Search report found citation-worthy pages tend to have.
• Update the page itself, not just the date stamp. A changed date on unchanged text is the kind of thing that erodes trust once anyone checks — do the actual refresh underneath it.

Overcorrections that backfire

Forcing every single heading into a question. Not every section is naturally a question, and a page where "What This Section Covers?" replaces "What This Section Covers" reads like it was run through a template, not written by someone who understands the topic. Convert the headings that genuinely improve as questions; leave the rest alone.

Treating one experiment as gospel. The whole point of this particular study is that a tactic surviving one test isn't proof — it's a hypothesis worth checking again. Apply the two findings that held up twice; hold the rest more loosely until more evidence exists.

Faking freshness instead of doing it. Bumping a "last updated" date without changing anything underneath it is exactly the shortcut this finding invites, and it's the one most likely to backfire the moment an AI system's answer turns out to be wrong because the content underneath the date hadn't changed at all.

Splitting topics so finely the page turns into a list of fragments. Two distinct ideas deserve two headings. Twelve one-sentence headings under one topic just fragments the argument and makes the page harder to read, for a machine or a person.

Reading a study like this once is easy. Going back through every page on a live site, heading by heading, and applying it — that's the part that quietly never happens on its own, which is why it's built into the ongoing work behind Get Found on Google and AI.

FAQ

What is GEO, and how is it different from AEO?

Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO) describe close to the same practice — getting a site cited by AI systems rather than only ranked by traditional search. The fuller breakdown of how SEO, AEO, and GEO fit together covers where the terms diverge; in day-to-day use, most people treat them interchangeably.

Should I rewrite every heading on my site as a question?

No. The finding is that headings phrased as questions performed better on average across two tests — not that every heading needs a question mark. Convert the ones that read naturally as something a customer would ask; leave descriptive labels alone where a question would sound forced.

How many citation events is 775, really?

It's a real sample for a field test — split across two separate campaigns, 15 keywords each, and four to six AI platforms — rather than a handful of screenshots. It's not a peer-reviewed study with a huge dataset, which is exactly why the second experiment testing the first one's conclusions again matters more than the citation count on its own.

Does this apply to a page that isn't a blog post — like a homepage or a service page?

Yes. The heading-phrasing and topic-separation findings apply to any page with structural headings, not just long-form articles. A service page with a heading like "Pricing" could just as easily read "How much does a website audit cost?" if that's genuinely the more natural phrasing for that section.

What does it mean that one conclusion didn't survive the second experiment?

It means at least one tactic that looked like a rule after the first campaign turned out to be specific to that campaign, not a general pattern. Search Engine Land didn't specify which one in the coverage available at publication — the useful takeaway is the discipline itself: test a finding more than once before treating it as settled.

Do I need to add a visible date to every page on my site?

Focus on pages where freshness plausibly matters to the answer — service pages, pricing-adjacent pages, anything time-sensitive. A static "About" page describing who you are doesn't need the same treatment as a page an AI system might cite for a "what does this cost right now" question.

Is this the same as the FAQ heading format DBO already uses?

Related, not identical. The ### Question? format in FAQ sections is a separate, older convention this site already follows for schema reasons. This finding is about the ordinary H2 and H3 headings in the body of a page — sections that were never part of an FAQ block to begin with.