TL;DR
- Trustworthy AEO advice sorts into four labels — peer-reviewed, independent practitioner, vendor-sponsored, and speculative — and most circulating articles wear the fourth without saying so.
- The only peer-reviewed AEO paper as of early 2026 is the GEO paper (Princeton, Georgia Tech, IIT Delhi, Allen AI) at KDD 2024, which reports up to a 40% boost in generative-engine visibility.
- Independent practitioner studies — Indig’s 18,012-citation analysis, SE Ranking’s 129,000-domain study — are where the field earns its keep. The test is whether the researcher profits from you believing the specific claim.
- Vendor-sponsored studies (HubSpot AEO launch, Frase, GenOptima) are fine for direction but not for magnitude — assume the real number is smaller than the decimal point suggests.
- Run four checks on any AEO article: does it cite numbers, can you follow one to its source, does the source sell the fix, and does it name a method and a sample.
Every AEO article you read this month probably sounded familiar.
Add markup. Write FAQs. Get cited. The advice rhymes from site to site.
You read three articles, close the tab, and still cannot tell what is real and what is marketing.
That feeling is not your fault. The field is young. Most of the writing about it comes from people selling something.
A small amount comes from researchers doing careful work. And a lot of the rest is well-meaning recycled SEO.
This piece is about how to tell the pieces apart. It gives you four simple labels to put on any AEO claim you meet. Once you have the labels, the noise shrinks.
Why AEO advice all sounds the same right now
The same few numbers show up in article after article. One piece claims a 3x citation uplift from FAQ markup. Another says 62% of searches are voice.
A third warns of 60% visibility loss by 2026 without the right markup. You can find those lines on ten different sites with ten different bylines. Nobody cites a primary source you can click.
There is a reason. The AEO content market has three large groups.
First, SEO agencies relabelling their old services with new words. Second, companies selling AEO tools and writing content that sends you back to their product. Third, a small number of researchers and practitioners publishing actual studies with named methods and named samples.
The first two groups are louder. The third group is where the real signal lives.
Different kinds of businesses see different shares of this content. I wrote about that in an earlier post on how AI search affects some businesses more than others. Local shops read different AEO advice from SaaS founders.
But the reader-side problem is the same. How do you tell which piece is worth acting on?
The four labels on the jar
Think of an AEO claim the way you think of food on a shelf.
You look at the label before you trust the contents. Peer-reviewed. Independent-practitioner.
Vendor-sponsored. Speculative. Four labels, four different kinds of trust.
No label means no trust, by default. A peer-reviewed label is rare and heavy. A vendor-sponsored label is useful in small doses. A speculative label means you treat the claim as a question, not an answer.
The rest of this article is a short tour of the four labels. When you finish, the next AEO article you read will sort itself into one of them within a paragraph.
Label one: peer-reviewed
This is the thinnest slice of the jar. Only one peer-reviewed paper sits at the centre of the AEO field as of early 2026.
It is called GEO: Generative Engine Optimization. The authors come from Princeton, Georgia Tech, IIT Delhi, and the Allen Institute for AI. It was published at KDD 2024, the main academic conference for this kind of work.
The paper tests a benchmark of real user queries. The headline finding is a boost of up to 40% in generative-engine visibility from specific content changes.
A peer-reviewed paper is heavy because it has a named method, named authors, and reviewers who do not work for the authors. You cannot buy that. When you find one, read it.
The GEO paper is short and in plain English. It sits at arxiv.org.
One paper is not a field. But one real paper is more signal than a hundred vendor blog posts stacked together.
Label two: independent practitioner studies
Most of the useful evidence in AEO lives here.
Independent practitioner means a researcher or analyst with a transparent method and a real sample. It also means they have nothing to sell on the specific claim they are making. Two examples stand out from the last six months.
Kevin Indig, an independent SEO analyst, published a February 2026 study on how AI decides which sources to cite. His team pulled 1.2 million search results and narrowed them to 18,012 verified ChatGPT citations. Then they measured where inside a page each citation came from. 44.2% of citations landed in the first 30% of the text. That is a hard, countable finding.
SE Ranking, a tool company, published a study in November 2025 on what predicts a ChatGPT citation. They analysed 129,000 domains and 216,524 pages across 20 industry niches. The single best predictor was the number of referring domains.
A distant second was domain traffic. Google ranking position came third.
That third-place rank finding is the bridge to why mid-ranking SEO sites win — Google’s middle of the index is precisely where AI engines find pages that cover one topic deeply.
Section structure mattered too. Pages with sections of 120 to 180 words averaged 4.6 citations. Pages with sections shorter than 50 words averaged 2.7.
SE Ranking sells SEO software, so they are not perfectly independent. But the claim they are publishing is that referring domains predict citations. That is not something SE Ranking sells a product to fix. The data is still worth reading.
That is the test. Does the researcher make money if you believe their specific claim? If the answer is no, the label sits closer to independent.
Label three: vendor-sponsored studies
The honest version of this label is that the company selling the solution is also measuring the problem. That is fine, as long as you read it knowing what it is.
HubSpot launched its AEO product on April 14, 2026. In the same announcement, HubSpot reported two first-party numbers. Organic traffic across its customer base is down 27% year on year.
Beta users of the new AEO tool saw AI referral traffic grow 20% against non-users. Both numbers are marked as HubSpot proprietary data.
This is not a dishonest study. HubSpot is a credible company. The data is probably real.
But the 20% uplift comes from customers who chose to join the beta. They are keen, they are already acting, and they are being measured by the vendor that needs the number to look good.
Read vendor data for direction, not magnitude. You can trust the shape. AI referral traffic is growing for people who work on AEO.
You cannot trust the decimal point. The real uplift may be 20%, or it may be 8%, or it may be 30%.
The same rule holds for every AEO tool study you see. The tool company measured it. Assume the real number is smaller.
Label four: speculative claims
The fourth label is the one most articles wear without saying so.
One article says 62% of searches will be voice by next year. Another says FAQ markup delivers a 3x citation boost. A third warns that small-business sites will lose 60% of organic traffic by 2026 if they ignore AEO.
You have seen versions of these claims. You cannot find the source in ten minutes of looking. The trail goes article, article, article, then a dead link, then nothing.
Speculative does not mean wrong. It means unverified. A claim with no traceable origin is a hypothesis.
You can use it as a prompt to ask questions. You cannot use it to make a decision.
When a claim is speculative, the only honest response is to hold it loosely. Look for a harder version of the same idea from a real sample before you act.
How to actually read an AEO article
Here is a short test you can run on any AEO piece in under two minutes.
- First, scan for numbers. If the article has none, it is opinion — fine, but weigh it accordingly.
- Second, follow one number to its source. If the trail ends at another blog post, the claim is speculative.
- Third, check who paid for the data. If the source sells a tool that fixes the thing the number describes, the claim is vendor-sponsored. Read for direction.
- Fourth, check for a named method or a named sample. If you find one, the claim is independent practitioner or peer-reviewed. That is the signal.
Most AEO articles fail the second step. That is useful information. It tells you the piece is not a decision tool — it is a conversation.
Other questions worth answering
Why do some writers use GEO or LLMO instead of answer-engine optimization?
Because branding drives the choice more than meaning at this stage. GEO came first in academic work, and the KDD 2024 paper from Princeton uses GEO directly. LLMO and GSO showed up later as alternatives. Wikipedia’s April 2026 entry says no consensus definition splits these terms from one another.
Did HubSpot share data on how leads from AI search convert against several channels?
Yes, HubSpot’s April 2026 launch announcement included a 3x conversion figure. Customers in the beta saw AI search leads convert about three times better than leads from rival channels. HubSpot also reported organic traffic across its customer base is down 27 percent year on year. Treat the conversion number as direction, not magnitude.
How much has ChatGPT use grown over the past year or so?
ChatGPT weekly active users climbed from about 300 million in December 2024 to roughly 900 million in February 2026. That is 3x growth in about fourteen months. Google’s share of search queries dropped from 87 percent to 78 percent during the same window. The shift is real even if any single forecast wobbles.
What do agencies typically charge for help with AI search visibility?
Pricing splits into three tiers. Entry-level retainers cost 500 to 2,500 dollars a month for schema work and monitoring. Mid-market work runs 2,000 to 8,000 dollars a month and adds content production. Enterprise plans start near 10,000 dollars a month, per 2026 agency pricing summaries.
What this means for the next article you read
You do not need a media-literacy degree to read AEO content well. You need four labels and a little patience.
Peer-reviewed evidence is rare and strong. Read it when you find it. Independent practitioner studies are where the field earns its keep. Read them first.
Vendor-sponsored studies tell you the shape of things. Read them knowing the magnitude is soft. Speculative claims are questions in disguise. Read them and keep moving.
The noise does not go away. But once the labels are on, the noise becomes sortable. You stop feeling behind. You start feeling picky.
Have you read an AEO article recently and felt unsure whether to act on it? That feeling is useful. The claim inside is probably vendor-sponsored or speculative. When that happens, more reading will not help.
You need a second pair of eyes. If you want mine, contact me. No pressure, no sales — just a calm look at what you are being told.