36 Businesses Scanned Across Five AI Assistants, and Half Were Nowhere to Be Found

Ask most owners whether their business shows up in AI answers and they'll open the assistant they happen to use, type their name, and read the result as a verdict. One check, and they call it settled. The trouble is that the assistants don't agree with each other, and the gap between what one shows and what the next shows is wide enough that a single look can leave an owner confident and wrong at once. Knowing where a business stands means measuring across all of them, and the picture that emerges is worse than most owners expect.

Fae Esparza is the Founder of Start Solutions AI, a firm that works on AI visibility for expert-led businesses, the problem of getting named when a customer asks an assistant for a recommendation. She came to it from an operations background, the kind that treats a claim as something to measure before it's repeated. Over the summer she did exactly that, running the same test on 36 businesses across five platforms and recording one fact for each: whether the assistant named them at all.

The results pushed her toward a warning she now repeats to anyone auditing their own presence. "Checking the one assistant you personally happen to use tells you very little," says Esparza. The scan showed why, and it started with how many businesses turned up nowhere.

What the summer scan found

Esparza ran the 36 businesses through ChatGPT, Gemini, Perplexity, Copilot, and Google AI Mode between late June and early August, noting for each whether the platform named it in an answer. Seventeen of them, 47 percent, were named by none of the five. "Not ranked low, not mentioned with a caveat," she says of that group. "Absent." Of the 19 that surfaced anywhere, six appeared on exactly one platform and only two appeared on all five.

No single assistant named businesses often. ChatGPT did it 33 percent of the time, Gemini and Copilot 28 percent each, Google AI Mode 22, and Perplexity 17. Esparza is careful about what the sample can carry: the businesses weren't chosen at random and skew toward owner-operated professional services in one metro area in the southern United States, and because the scans ran across seven weeks rather than a single day, some of what looks like disagreement between platforms could be movement over time. Her next wave runs all 36 in one day to pull those two effects apart. Her platform-level findings are written up in full.

Different platforms, different sources

The five landed on similar hit rates, close enough that you'd assume they were reading the same underlying picture and simply setting the bar in different places. If that were the case, the same businesses would keep surfacing across all of them. They didn't. A business that appeared on one platform was often missing from the next, and that mismatch is the tell. "Similar rates with poor overlap points at different sources, not different strictness," says Esparza.

That changes what an audit has to look like. If the assistants drew from one shared index, checking any one of them would stand in for the rest. Because they're pulling from different places, a clean result on ChatGPT says nothing about whether Gemini or Perplexity has ever heard of a business, which is how an owner ends up reassured by a test that only covered a fifth of the field.

Cited, then passed over

The scan measured presence, but Esparza points to a second gap that a presence count misses. She has audited a business that was the single most-cited source in an AI answer and still showed up in none of the buying recommendations that followed. The assistant leaned on their website to explain the topic, then sent the buyer to a competitor. "Being cited and being recommended are two different states, and people collapse them," says Esparza.

For an owner, the two can feel like the same win. They pull apart quickly. Getting quoted means the model trusts your page enough to build an answer from it. Getting recommended means it trusts you enough to send someone your way, and that second kind of trust leans on signals that live off your own site. It's the same dynamic behind off-site trust signals: the businesses an assistant hands to a buyer tend to be the ones other places vouch for, not only the ones that explain a topic well.

Cleaning up what came before

When Esparza takes on a client, the first pass is rarely flattering to whoever came before, even when the prior work was solid. "Even if they have years and years of attention toward their SEO, I almost always find missing links, things that need to be cleaned up," says Esparza. The messiest accounts are the ones that hopped between vendors, each leaving its own half-finished layer, and those turn into heavy pruning before any progress shows.

She's deliberate about how she delivers that news, especially to an owner loyal to a long-time agency. "I would never tell a client, 'for 15 years you've been using the same agency and it's all wrong,'" she says. Instead she lets the baseline carry it, showing plainly what needs to be elevated and what needs to be tidied, which turns the audit into a conversation about where the company wants to go rather than a verdict on where it's been. That judgment about what's off is the kind of read machines still miss.

Making a brand legible

The response Esparza builds on top of the diagnosis is less exotic than the numbers suggest. AI visibility starts from the search work a brand already has, which is why she likes inheriting a client with clean SEO to build on. What she adds on top is coherence, making a brand legible enough that an assistant can place it, because one that reads as noise gets left out of the answer. "A confused brand, or confused SEO, that's what makes AI skip you and say, 'I don't understand what's going on,'" says Esparza.

That's why she runs a baseline before she'll talk strategy. Most of the businesses she takes on have paid for years of search work without anyone measuring what the assistants say about them, which leaves nothing to steer by. "How do you know what's going to work if you don't have a baseline?" she says. Her approach is to tie effort to outcomes first, then tidy the entities and connections until the brand explains itself the same way to a person and a machine.