Key Takeaways

  • Foundation Inc.'s State of B2B AI Discovery research found ChatGPT and Google AI Overview cite almost entirely different sources for the same enterprise software queries, overlapping mostly on vendor pages themselves.
  • AEO is not one discipline. Optimizing for ChatGPT citations and optimizing for AI Overview citations are two separate motions with two separate source universes.
  • Most AI SEO tools on the market today measure visibility across platforms, but very few tell you which corner of the internet each engine is actually pulling from, which is the more useful question.
  • The practical response for a CMO is to run parallel content and PR programs: one aimed at the sources ChatGPT trusts, one aimed at the sources Google's AI surface trusts, with your owned pages as the shared foundation.
  • Skip this split if your buyers still convert almost entirely through classic organic search. The two-playbook approach is a bet on where discovery is heading, not where every dollar sits today.

A piece of research came across my desk last week that I haven't been able to put down. Foundation Inc. ran the same enterprise software queries through ChatGPT and Google's AI Overview and looked at what each one cited. The overlap was almost nothing. The only sources both engines consistently pointed at were the vendor's own pages, and everything else, the third-party context that shapes how a buyer actually forms an opinion, came from completely different corners of the web.

I ran a version of the exercise myself to make sure I believed it. Asked both engines a plain buyer question about data catalogs for a mid-market company. ChatGPT came back leaning on practitioner blog posts, a couple of long Reddit threads, and one opinionated Substack from someone I've read for years. AI Overview came back with a tidy stack of category roundups from the SEO-heavy publications that have been ranking on that query for years, plus the two biggest vendors' own comparison pages. The only source that showed up in both answers was the vendor site itself.

That's the whole finding.

If you're a CMO who has been building an AEO program this year, that should change what you do on Monday. Most of the AI SEO tools we've been buying, and most of the internal conversations we've been having, treat AI visibility as one thing. It's at least two things, and the sources feeding them barely know each other exist.

What Foundation actually found

The Foundation Inc. study is worth reading in full, but the headline is this: across a set of enterprise software queries, ChatGPT and Google AI Overview cited mostly different sources, and the shared citations were dominated by the vendors' own websites. Beyond that thin overlap, each engine pulled from its own preferred neighborhoods of the internet.

That's a bigger deal than it sounds. It means the third-party pages that make a buyer trust a vendor, the reviews, the roundups, the analyst-adjacent posts, the community threads, are not shared inventory. If you got mentioned in a piece that ChatGPT loves, there's no reason to assume Google's AI surface has even indexed the context around that mention, let alone chosen to cite it.

The reason this matters isn't the specific list of sources, which will shift as both engines evolve. The durable insight is that the retrieval layer underneath each engine has been built differently. ChatGPT and AI Overview are making different editorial choices about who to trust, and those choices are stable enough that you can plan against them.

Why the two engines don't agree

The mechanics are fairly simple once you say them out loud. Google's AI Overview sits on top of Google's index and Google's ranking signals, which have been tuned for two and a half decades around a particular idea of authority: backlinks, freshness, entity coherence, all the usual furniture. When AI Overview picks sources, it's picking from a pool Google was already prepared to rank.

ChatGPT's retrieval is a different animal. It leans on its own search partnerships, its own crawl, and a set of preferences that reward long-form explainers, structured comparisons, and content from sources that show up repeatedly as reference material in the kind of text the model was trained on. Its taste is genuinely its own, and it's stubborn about it.

So when you ask both engines "what's the best data catalog for a mid-market company," they're not just re-ranking the same ten pages in a different order. They're often looking at different ten pages entirely, and the only thing they reliably agree on is that the vendor's own site should be in the mix.

What this does to your AEO strategy

The tempting response is to buy a tool that tracks visibility across both platforms and call it a day. There are plenty of options; the recent roundups from One Little Web, Freddie Chatterton, and Behind Rankings each list fifteen to twenty of them, and most of the credible ones (Semrush, Surfer, Rankscale, seoClarity per Zapier's roundup) now include some form of AI visibility tracking.

Tracking is table stakes; it tells you the score. The Foundation research is telling you the game has changed. You need two content and PR programs running in parallel, each aimed at a different retrieval layer, with your owned pages as the shared foundation both engines already agree on.

Track one: the ChatGPT corner

For ChatGPT, the target is the kind of source that shows up as reference material inside long, structured explainers. Practitioner blogs with a clear point of view. Category comparison posts written by people who have actually used the tools. Long threads on Reddit and Hacker News where a specific claim gets argued through. Substacks and personal sites from named experts in your category.

Getting cited here isn't a PR motion in the traditional sense. It's closer to a community motion. You want your product and your point of view to show up inside the artifacts these communities produce, which means being useful to the people who write them long before you need anything from them.

Track two: the AI Overview corner

For AI Overview, the game looks more like classic SEO with a twist. Google is still Google. Authority signals still matter, entity clarity still matters, and the sources AI Overview cites tend to be the ones Google was already ready to rank on the underlying query. The twist is that AI Overview seems to prefer sources that answer the question cleanly and structurally, with the answer near the top and the supporting evidence laid out underneath.

Which means for this track, the work is more familiar: strong on-page structure, clear H2s that mirror real questions, tables and lists where they earn their keep, and enough domain authority behind the page that Google trusts it as a citable source in the first place.

What to actually do about it

The move I'd recommend is to stop treating AEO as a single workstream in the plan and split it into two: one backlog for ChatGPT-shaped content and outreach, one backlog for AI Overview-shaped content and outreach, and a shared foundation of vendor pages that both engines already agree are the source of truth for the brand.

What matters isn't the split itself, it's what the split does to prioritization. Once you accept that the two retrieval layers are different, you stop arguing about whether a given piece of content is "good for AI" and start asking which AI. The average visibility number across a single dashboard stops meaning much. You staff against the two motions differently, because they genuinely require different muscles: community and relationships on one side, structured content and authority on the other.

When not to do this

The honest caveat: not every B2B company should be splitting their AEO program in two right now. If your pipeline still comes almost entirely from classic organic, paid, and outbound, and your buyers aren't yet starting their research inside ChatGPT or leaning on AI Overview to shortlist vendors, a two-track program is premature. You'll spend real money chasing a channel that isn't converting yet.

The tell that it's time is when you start hearing prospects say, in discovery calls, that they asked ChatGPT about your category and got a specific list back. Once that happens twice in a quarter, the split is worth the investment. Before that, keep your classic SEO program healthy and run one or two AEO experiments to build the muscle.

What's still unresolved

Six months ago I would have told you that AEO was mostly a rebrand of SEO with some prompt-flavored garnish on top, and that a good technical SEO program would carry you through the AI transition with minor adjustments. The Foundation data is what changed my mind. It's the first piece of evidence I've seen that says the retrieval layers are structurally different, not cosmetically different, and that a single program aimed at "AI" is going to underperform a program that treats ChatGPT and AI Overview as separate surfaces with separate source pools.

The question I don't have a clean answer to yet is how stable this split will be. If ChatGPT starts leaning harder on Google's index, or if Google's AI surface starts pulling from the community sources ChatGPT prefers, the two lists could converge. My read is that they won't converge fully, because the two companies have different strategic reasons for keeping their retrieval distinct. But it's worth watching, and it's worth keeping your program flexible enough to shift the ratio as the picture changes.

Frequently Asked Questions

What are AI SEO tools actually good for right now?

They're good for measurement and for surfacing which prompts your brand shows up in. Most of the credible ones, Semrush, Surfer, Rankscale, Search Atlas, seoClarity, do a reasonable job tracking visibility across ChatGPT, Perplexity, and Google's AI surfaces. What they're less good at is telling you which specific third-party sources each engine is pulling from for your category, which is the more useful question in light of the Foundation research.

Do I need a separate AEO program for every AI engine?

Not every engine, but at minimum two: one aimed at ChatGPT-style retrieval and one aimed at Google AI Overview. Perplexity and Claude tend to pull from sources that overlap meaningfully with ChatGPT's preferences, so a well-run ChatGPT track often gets you partial credit on those. Google AI Overview is the one that behaves genuinely differently and warrants its own workstream.

How does this change what my content team should be writing?

Less consolidation into one mega-page per topic, more variation. The ChatGPT track rewards distinctive, opinionated, community-flavored content that shows up as reference material inside longer explainers. The AI Overview track rewards structured, clearly-formatted pages that answer specific questions near the top and carry enough Google-side authority to be trusted. Same topic, two different artifacts.

Is classic SEO dead?

No, and treating AEO as a replacement for SEO is the fastest way to hurt your pipeline. AI Overview sits on top of Google's index, so classic SEO fundamentals are what get you into that engine's citation pool in the first place. The two disciplines are additive right now, not substitutive.

Where should a CMO who's new to this start?

Run the same twenty buyer-intent prompts through ChatGPT and Google AI Overview yourself, once a week for a month. Write down which sources each one cites. You'll see the split the Foundation research describes inside your own category within a few sessions, and that direct exposure is worth more than any dashboard for building the intuition you need to prioritize the work.

Sources