Why Page One Rankings Don't Guarantee AI Citations

Intro

In July I ran a small test that changed how I brief clients.

I took a software vendor that ranks sixth on page one of Google for a term doing about 5,400 searches a month. Solid position, competitive query, the kind of ranking most teams would be pleased with. Then I asked ChatGPT and Perplexity the question a buyer actually types when they are close to choosing.

Perplexity cited the vendor. ChatGPT named six competitors and two official portals, and did not mention them once.

Same company, same week, same query intent. One assistant treated them as a category player and the other did not know they existed.

Why Rank and Citation Have Come Apart

The two systems are not reading the same web.

Perplexity leans heavily on what Google has indexed, so a page-one ranking tends to carry over. ChatGPT assembles from a narrower set it has decided to trust: category comparison sites, official sources, community discussion, publications with visible editorial standing. Rank is one input among several there.

Which means the thing most SEO teams optimise for and the thing that increasingly decides whether a buyer hears your name have started to diverge. Two assets now, earned separately.

What the Sites Getting Cited Have in Common

This year my team scored close to 2,000 company websites in a single sector on what they publish and how it converts, and ran a parallel read of about 1,000 marketing job posts. The full study covers the conversion side. The authority findings are the ones relevant here.

Answer engines lean on demonstrable expertise. Named authors, stated credentials, clear editorial ownership, and above all original data that exists nowhere else. On those measures the field is thin:

• Original research or proprietary data on about 29% of sites
• Author bios carrying credentials on about 26%
• Dedicated author profile pages on about 5%
• A stated editorial policy on roughly 3%

Rolled up, around one in seven companies shows strong expertise signals of any kind.

That is a low bar, and it is worth sitting with. When an assistant assembles an answer about your category, it is choosing from whoever looks like a credible source. In most categories that shortlist is short because almost nobody has tried to get on it.

The Hiring Gap Behind the Visibility Gap

The job-post data explains why the field stays thin.

Answer-engine optimisation appeared in about one in eighty of the marketing roles this sector was hiring for. The same sector named TikTok in job adverts roughly ten times more often.

So the channel now mediating discovery is the one almost nobody is staffed to compete in. If you have been waiting for a reason to move early on AEO, that ratio is it. The capability gap is wider than the technical one.

What Actually Earns a Citation

The uncomfortable answer is that you have to be the source rather than a summary of one.

An assistant asked about your category will not cite your take on someone else’s survey. It cites the survey. So the asset that works is the one only you could publish: your own numbers, your own sample, your own measured result. A benchmark on a boring, low-volume term with no incumbent will outperform a well-optimised opinion piece on a competitive one, because there is nothing else to cite.

That is also why original research has become the cheapest authority play available. Most companies already hold the raw material. Support tickets, win-loss notes, pricing data, onboarding times, whatever you measure to run the business. Very little of it ever gets published, which is exactly why publishing it works.

The second thing that helps is mundane. Put a real name, a real credential, and a real bio on what you publish, and state who is editorially responsible for the site. Three per cent of the sector has done the last one. It takes an afternoon.

Structure matters more than it used to as well, though less than people hope. An assistant extracting a claim needs to find it cleanly: the figure stated in a sentence rather than trapped in a chart image, the method described in plain words, the date visible, the sample size named. I have watched a genuinely original stat go uncited for months because it only existed inside a PDF and a graphic.

None of that creates authority on its own. It removes the friction between having something worth citing and being citable, which is a different job and a much smaller one.

How to Check Your Own Position

You can run the test I ran in about fifteen minutes and it costs nothing.

Write down the three questions a buyer asks in the week before they choose. Not your head terms, the actual questions. Put each one to ChatGPT, Perplexity, and Gemini in a fresh session with no personalisation. Note whether you appear, who does, and which pages get cited.

Then compare that list against your Google rankings for the same intent. Where the two lists disagree, you have found the gap that your rank reporting is not showing you.

I expected that gap to be narrow when I started testing. In the case above the vendor was sixth on Google and entirely absent from one of the two assistants a buyer is most likely to ask. That is not a ranking problem, and no amount of on-page work fixes it.