The Consensus Gap

Enhance your expertise with Progress Memo’s weekly professional insights. Subscribe for free!

Most groups speak about “AI visibility” prefer it’s one factor. New information on 3.7 million citations throughout ChatGPT, Perplexity, and Google AI Overviews suggests it isn’t. And the hole between the three engines is wider (and extra strategically essential) than your dashboard probably admits.

Immediately’s memo breaks down:

• Why a blended AEO rating hides the one discovering that issues.
• Which web page sorts and domains truly journey throughout engines.
• The shift from measuring AI presence to measuring portability.

One of many greatest variations between AEO and search engine marketing is that AEO plays on more platforms.

Omnia information reveals throughout a number of samples that solely 2.35% to 2.45% of cited URLs appeared in ChatGPT, Perplexity, and Google AI Overviews for a similar immediate. 91% of citations appeared in just one engine.

Conclusion: AI visibility will not be a single leaderboard. As an alternative, it’s three totally different distribution programs that typically overlap and often don’t.

Solely 2% Of URLs Get Cited By All 3 Engines

Most individuals would guess that if a URL will get cited by one main AI engine, it has an affordable shot at showing within the others.

However the 20,000 immediate pattern reveals solely 2.37% of cited URLs present up throughout all three engines for a similar immediate.

In the meantime, 91.07% present up in just one. These two numbers belong subsequent to one another as a result of they clarify one another. The remaining ~7% overlap in pairs, which implies engines are drawing from largely disjoint swimming pools quite than rating the identical pool in a different way.

For AEO/search engine marketing groups, meaning a single composite visibility rating is the incorrect unit of measurement. Averaged AEO scores conceal this. A model can look sturdy in mixture and be invisible in 2 of three engines. Groups chasing one blended AI visibility quantity are compressing three ranking systems into one metric and calling it technique.

The two% Holds Throughout Each Minimize

The ~2% overlap charge and ~91% unique charge keep nearly completely flat throughout 4 samples.

That consistency issues greater than the precise decimal level. The consensus hole will not be an artifact of 1 question set or one time window. It seems structural.

In Q3 2025, common overlap was 2.2%. In This autumn 2025 and Q1 2026, it rose to 2.7%. Engine-exclusive citations fell from 90.1% to about 88%. So sure, a small quantity of convergence. However even after that shift, fragmentation nonetheless dominates.

Industrial Prompts Don’t Converge Both

The intent break up is without doubt one of the quietest however most helpful components of the dataset. You possibly can argue that industrial queries ought to produce extra consensus. When somebody searches for [the best CRM], [best running shoes], or [best project management software], the pool of acceptable sources feels narrower than it does for broad informational prompts.

Surprisingly, the information doesn’t help a giant distinction.

Industrial prompts present 2.4% common overlap. Informational prompts present 2.0%. Even when the question ought to slender the reply set, the engines nonetheless select totally different sources more often than not.

That pushes in opposition to a standard intuition in search engine marketing and content material technique. Groups typically assume high-intent queries are the place shared authority will present up. The alternative seems nearer to true. Even in industrial territory, every engine’s personal retrieval logic, what sources it trusts, what codecs it prefers, is doing many of the work.

Guides Beat Homepages By 2x

The web page kind breakdown under reveals guides and tutorials have the best cross-engine overlap at 2.3%, adopted by blogs at 1.8%, class pages at 1.6%, product pages at 1.2%, and homepages at 1.1%.

Two classes:

1. First, explanatory content material travels higher than model or transactional property. If you would like one of the best shot at exhibiting up throughout engines, the strongest candidate will not be the homepage and never the product web page. It’s the web page that helps, explains, compares, or teaches, however take into account that these are additionally content material codecs that AIs can reply straight properly.
2. Second, even one of the best web page sorts carry out badly in absolute phrases. Guides will not be successful throughout engines in any significant sense. The proper learn on this isn’t “publish extra guides and you’ll win all over the place.” It’s easier than that: Helpful content travels better than brand content.

Visibility Is Not The Identical As Portability

One of many best errors on this house is to confuse citation frequency with citation portability. Wikipedia is the cleanest instance. It seems 16,073 occasions within the dataset, however just one.3% of these appearances are common throughout engines. Reddit seems 14,267 occasions, however solely 0.1% are common. Reuters reveals up 1,202 occasions and nonetheless lands at 0.0% common overlap.

That’s the reason an essential metric is portability. A site can present up throughout one engine and barely journey, which implies a model trying dominant in an mixture dashboard could also be one platform’s behavior away from invisibility. Presence tells you whether or not you might be seen. Portability tells you whether or not that visibility is resilient.

What This Means For Operators

The sensible implication is easy: Cease treating AI visibility as one factor. Study the excellent visibility of your area by measuring:

1. Presence, the % of your tracked prompts the place your area seems in any engine. Presence tells you whether or not you’re seen.

2. Portability, the % of your cited URLs that seem in all three engines. Portability tells you whether or not that visibility is resilient.

3. Focus, the % of your citations that come from a single engine. Focus tells you which ones engine your present dashboard is secretly constructed on.

If the overlap between engines is that this low, a single AEO technique is just too summary to be helpful.

Once we method AI visibility from a holistic perspective, it forces sharper questions:

• Which engine issues most for us?
• Which of our property journey throughout engines, and which solely work in a single?
• Are we measuring presence after we needs to be measuring portability?

This additionally modifications how model groups ought to take into consideration diagnostics. A weak homepage throughout engines is probably not a homepage downside. It’s a symptom of one thing broader: Engines favor utility over model centrality. In that world, visibility comes much less from being the official supply and extra from being the helpful supply.

The strategic query is not, “How can we rank in AI?” We must always as an alternative be asking ourselves, “How can we construct property that survive totally different engine preferences?” That could be a narrower query. It is usually a greater one.

Methodology

There are just a few caveats to this evaluation:

• The dataset is skewed towards Omnia’s buyer base.
• The intent and page-type cuts depend on regex classification, which is beneficial for directional evaluation however not excellent taxonomy work.

These caveats don’t weaken the principle discovering a lot. The most important sign will not be precision on the edges. It’s consistency on the middle. Irrespective of how the cuts change, the identical sample resurfaces: little or no overlap, very excessive engine-specificity, and solely modest variations by time, intent, or web page kind.

Dataset Measurement And Time Window

The evaluation attracts on 4 immediate samples. Three cohorts of 5,000 prompts every, tracked from Jan. 1, 2025; July 1, 2025; and Jan. 1, 2026. A separate 20,000-prompt random pattern underpins the headline 2.37% and 91.07% figures. The time-view minimize spans Q3 2025 by means of Q1 2026 (thus far) and covers 3.7 million URL citations in complete. Industrial/Informational/Different intent splits are drawn from roughly 2.6 million URLs throughout the mixed pattern. Web page-type splits span 4.1 million URL appearances.

How Prompts Have been Chosen

The 20,000 prompts are drawn as a random pattern from Omnia’s reside immediate monitoring pool. The pool displays what actual advertising groups selected to trace, weighted towards Omnia’s buyer geography (Spain-heavy, plus UK, Nordics, and different EU markets). Every immediate runs in its nation’s main language, so Spanish is overrepresented versus a U.S.-only dataset. Business combine is fintech/insurtech, journey, SaaS, B2B companies. Deal with findings as directional for European AI search.

Engine Protection

The research covers three engines: ChatGPT, Perplexity, and Google AI Overviews. Every fires the identical immediate concurrently inside the identical minute, twice a day, with nation localization, and every engine queried in its default web-enabled, unauthenticated state. Perplexity monitoring runs on Sonar, whereas ChatGPT and Google AI Overviews use every vendor’s default manufacturing mannequin for logged-out net shopping (which neither OpenAI nor Google pins publicly to a selected model).

Classification Methodology

Intent and web page kind are assigned by regex. Intent buckets are Industrial, Informational, and Different. Web page-type buckets are Information/tutorial, Article/weblog, Class web page, Product web page, Homepage, Wikipedia, and Different. The principles are keyword- and URL-pattern-based, which makes them quick sufficient for a multi-million-URL dataset however coarse on the edges. Edge instances fall into Different, which is why Different carries a excessive share in each the intent and page-type tables. Deal with the regex cuts as directional, not authoritative.

Extra Assets:

Featured Picture: FGC/Shutterstock; Paulo Bobita/Search Engine Journal