Google AI Mode Is Not the AI Search You've Been Optimizing For

When businesses ask about AI visibility, they usually mean one thing: are they showing up in ChatGPT, Perplexity, Gemini, Claude? Those platforms answer direct recommendation queries. They're the ones someone turns to when they type "best dentist in Austin" or "who should I hire for a kitchen renovation."

Google AI Mode is different. It launched in conversational form earlier this year and is now rolling out to all Google users. It looks like AI search, it feels like AI search, and it appears in the same place people have used Google for 25 years. But the signals that determine who appears in AI Mode are almost entirely separate from the signals that determine who gets cited by ChatGPT or Perplexity. Treating them as the same problem leads to months of work aimed at the wrong surface.

Three AI surfaces, not one

In our May 23, 2026 investigation into whether to add Google AI Mode to Sourcepull's audit scope, we pulled Ahrefs' analysis of 540,000 query pairs across AI Mode and AI Overviews. The URL overlap between the two surfaces: 13.7%.

That's not a small gap. It means 86.3% of the URLs appearing in AI Mode answers are different from the URLs appearing in AI Overviews answers, even though both are Google products running on the same underlying search infrastructure.

AI Overviews already has significant divergence from ChatGPT and Perplexity. Add AI Mode into the picture and you have three largely distinct citation surfaces. A business can be:

- Well-cited by ChatGPT (strong entity data, good directory presence, consistent external descriptions) - Appearing in AI Overviews (different content signals, tied more closely to organic ranking) - Missing from AI Mode entirely (because GBP is incomplete and review recency is low)

Or any other combination. Knowing you have an AI visibility problem doesn't tell you which surface you're failing on, and that's the only question that connects to a fix.

The signal sets are mostly non-overlapping

Our May 2026 scope analysis documents what actually moves AI Mode results for local businesses. The primary signals break down roughly as follows: Google Business Profile quality carries around 32% of the weighting -- photos, hours accuracy, category selection, post recency. Review signals account for approximately 20% -- not just star ratings, but review quantity, review recency within the past 90 days, and whether the business responds to reviews. NAP consistency across directories makes up roughly 13%.

The LLM citation signal set is different. Our May 23, 2026 review of the Ahrefs controlled schema study and the subsequent practitioner commentary identified the actual drivers of ChatGPT and Perplexity citation frequency: directory presence in the specific platforms each AI trusts for the relevant category, topical authority expressed through content covering the full range of sub-queries a searcher might ask, and entity consistency -- the same description of your business appearing the same way across multiple external sources.

These are nearly non-overlapping lists. A business with a flawless GBP, 400 recent reviews, and a 4.9 average rating can still score 2/10 on Perplexity if it has no indexed external presence and zero directory listings. ChatGPT's training pipeline and Perplexity's retrieval index don't read your GBP the way AI Mode does. The channels are architecturally separate.

Running the other direction: a business with a tightly built entity graph, consistent listings across 12 directories, and high Perplexity citation rates might be invisible in AI Mode because no one on the team has logged into Google Business Profile in 14 months. The fix plans for each case have almost nothing in common.

Schema doesn't bridge the two worlds

The signal that feels most transferable is schema markup. It's in every GEO checklist, every local SEO guide, and most AEO recommendations. It seems like the kind of infrastructure improvement that would help across all AI surfaces.

Our May 23, 2026 methodology rec documents the first controlled schema study (Ahrefs, May 11, 2026 -- 1,885 pages that added JSON-LD schema, matched against 4,000 control pages, difference-in-differences analysis):

| Platform | Citation change after schema | Significance | |---|---|---| | Google AI Mode | +2.4% | Not statistically significant | | ChatGPT | +2.2% | Not statistically significant | | Google AI Overviews | -4.6% | Statistically significant, negative |

No meaningful uplift on any surface. The Ahrefs study's critical limitation: every page in the dataset already had 100 or more AI citations before schema was added. The study tests whether schema pushes already-visible pages to be cited more. It doesn't test whether schema helps uncited pages get noticed. But for businesses with any existing AI presence, schema addition isn't moving citation frequency.

The right frame for schema is entity registration, not visibility boost. LocalBusiness schema with a consistent name, address, phone, and sameAs links to verified directories tells AI systems what your business is and where it operates. That's a prerequisite for accurate citation, not a driver of frequent citation. It's worth implementing for that reason -- but it doesn't solve an AI Mode problem or a ChatGPT problem. It sets a baseline that both problems build on.

The one signal that does transfer between AI Mode and LLM platforms is NAP consistency -- identical name, address, and phone number across your website, your GBP, and every directory where you're listed. AI Mode uses NAP as part of its citation and NAP signals cluster. LLM platforms use it for entity resolution. Inconsistent NAP creates entity fragmentation on the LLM side and citation instability on the AI Mode side. Fixing it is the closest thing to a universal improvement across all surfaces.

Knowing which failure you have changes what you do first

Our May 24, 2026 methodology work on the selection-versus-absorption framework adds one more layer to this. Even when a business appears in AI answers, there are two distinct outcomes: citation selection (the AI includes the business in its response) and citation absorption (the AI actually uses accurate information about the business to shape what it says).

A business can be selected by Perplexity but absorbed poorly -- appearing in the answer but with a wrong address, a mismatched service description, or a phone number from a closed location. That's the misattribution case, and the fix is correcting bad data in the specific external sources the platform is pulling from. Adding more directory listings won't help; fixing the data in the listings that already exist will.

The same distinction applies in AI Mode. A business can appear in an AI Mode result card with incorrect hours, a wrong category, or a description pulled from an outdated GBP photo caption. The fix there is GBP correction, not entity graph work.

Both citation failure modes -- absent entirely, or present but inaccurate -- require knowing where in the pipeline the failure is happening. That diagnosis is different for AI Mode than for LLM platforms.

What this means in practice

Sourcepull's audit covers the LLM citation layer: ChatGPT, Perplexity, Gemini, and Claude. Our May 2026 scope analysis concluded that Google AI Mode should not be in Sourcepull's audit scope for now -- AI Mode's primary fix actions involve GBP optimization and review management, which are Google SEO problems rather than LLM infrastructure problems. Adding AI Mode to an audit that doesn't connect to a GBP-layer fix plan would produce a gap report without actionable next steps.

If someone in a sales conversation asks whether Sourcepull covers Google AI Mode, the honest answer is: no, and here's why the distinction matters. AI Mode visibility is a GBP problem. LLM citation visibility is an entity infrastructure problem. They're both worth solving, but they require different vendors, different timelines, and different fix plans.

The first question worth answering is which problem you actually have. If your business has been showing up reliably in Google local results but buyers keep saying they "found someone else through AI search," the gap is likely the LLM citation layer -- not AI Mode. If Google's own results for your category consistently surface competitors above you, the AI Mode problem probably follows from the same underlying GBP and review gap.

Signal Check at sourcepull.ca runs 40 live queries across ChatGPT, Perplexity, Gemini, and Claude, scores by platform, and identifies whether the gap is footprint (you're absent because you have no indexed presence), accuracy (you appear but with wrong information), or something structural like a crawler block. That tells you where to start on the LLM side. The GBP side has its own audit tools; the two investigations don't need to happen in the same place.