In 2025, 82% of consumers found AI more helpful than traditional search. In 2026, that number is 54%. Usage went the other way. Fractl found that 70% of consumers said their AI use for search increased over the past year, while only 4% said they have never used AI tools for search. So people are using AI search more and rating it lower. That looks like a contradiction, but it makes sense once you look at what people do after they get an AI answer: they check it. That checking step now shapes how people find businesses and information online.

What the Fractl survey actually measured

The numbers come from Fractl's 2026 AI search study. Fractl surveyed 1,008 U.S. consumers and 150 marketers in Q2 2026. Coverage of the study used two different math framings, so here's the clear version. The drop from 82% to 54% is a 28-point drop in percentage points. Some outlets reported it as a roughly 34% decline, which is the same change measured as a share of the starting number.

The bigger change is in the skeptic group. The share of consumers who rate AI as less helpful than traditional search rose from 3% to 17%. Most people haven't turned against AI search. They've gone from excited to cautious.

The age breakdown isn't what most people would guess. Baby Boomers (63%) were more likely than Gen Z (47%) to say AI is more helpful than traditional search. One likely reason is that heavier users run into more of AI's mistakes, so they've learned not to take answers at face value.

One caveat: this survey asked whether AI is more *helpful* than search. It didn't measure trust directly. Many headlines treated "helpful" and "trust" as the same thing. They're related, but they aren't identical.

Trusting AI and checking it aren't opposites

Another survey shows why the gap matters. Yext's 2026 Consumer Search Behaviors Report asked people what they did after an AI tool recommended a local business. Yext surveyed 1,120 US adults in March 2026, and most of them said they trust AI. 74% of AI users rate their trust in AI recommendations at 4 or 5 out of 5. And yet: 93%+ still take at least one verification step before acting.

Here's what that checking looks like: after receiving a recommendation, 62% immediately search Google for more information, 58% visit the business website directly, and 52% click through to sources cited within the AI's response. The people who trust AI the most still check. High-trust users — those who rate their confidence in AI at 5 out of 5 — verify via Google at 62.0%. Neutral users verify at 63.1%.

So trust in AI doesn't mean people skip the checking. They treat the AI answer as a shortlist, then confirm it somewhere else before they act. Only 6.7% of AI users act on a recommendation without doing any additional research.

Why people started double-checking

Neither survey directly explains why perceived helpfulness fell. But the data points to a few likely causes.

AI answers sometimes turn out to be wrong. People who use AI every day eventually get a wrong store hours listing, a product that doesn't exist, or a confident summary that leaves out the important caveat. After that happens once or twice, checking becomes a habit.

People are wary of brands that lean heavily on AI. Fractl found the share who say heavy AI use by a brand makes them trust that brand *less* nearly doubled, from 20% to 39% year over year, and hits 54% among Gen Z.

People want AI content labeled, and most brands don't label it. 84–91% of consumers want AI-assisted content clearly labeled, but only 20% of brands consistently disclose it. When people can't tell what was written by AI, they're more likely to doubt all of it.

For high-stakes topics, people still go to regular search. Yext's data shows 57% of users still prefer traditional search for important YMYL (Your Money Your Life) topics, like health and finance. The bigger the decision, the more likely people are to look past the chatbot.

How the checking step works

Think of it as a loop. AI suggests some options, and the person then goes to two or three other places to confirm them. Based on the Yext data, those places are, roughly in order:

  1. A Google search for your name. It's the most common check. People want to see your reviews, your address, and whether you look like a real, established business.
  2. Your own website. This is the second most common step. If your site has old prices, broken pages, or nothing that matches what the AI said, you can lose the visitor here.
  3. The sources the AI cited. About half of AI users click them. If one of those citations points to your site, the page it links to has to back up the claim.
  4. Reviews. After an AI recommendation, star rating is the number one purchase influencer after AI (34%), followed closely by word of mouth (30%), review recency (29%), review sentiment (28%), and review count (28%).

Being mentioned in an AI answer gets you onto the shortlist. Whether people actually choose you depends on what they find when they check.

What to do with this

If you run a site or a business: Search Google for your business name and read the results the way a doubtful stranger would. Make sure your hours, prices, and product details match across your site, your Google Business Profile, and the review sites. If they don't match, AI may repeat the mistake, and people checking will notice. Make your key facts easy to quote and easy to confirm: dated numbers, named sources, and specific claims instead of vague ones. We covered how to structure pages for this in 7 ways to make your content easy for ChatGPT and Google AI to cite. Once people are checking AI answers, the same advice also helps you pass that check.

If you're the one searching: Check the way most people already do, just more deliberately. Use the AI for the first pass, then click at least one cited source to confirm the claim is really there. For money, health, or legal questions, go to a primary source. Tools like Perplexity that show sources inline make this faster, but read the source yourself. A citation alone doesn't mean the source says what the AI claims.

For everyone: Don't read the drop from 82% to 54% as AI search failing. The early excitement wore off, and people now treat AI answers more like a recommendation from an acquaintance: useful, worth hearing, and worth checking before you act on it.

FAQ

Is trust in AI search actually falling in 2026?

Perceived helpfulness is falling. In Fractl's Q2 2026 survey, the share of U.S. consumers who said AI search is more helpful than traditional search fell from 82% to 54%, and the share rating it less helpful rose from 3% to 17%. Usage kept growing over the same period, with 70% saying they use AI for search more than a year earlier.

What do people do after an AI tool recommends something?

Most of them check it. In Yext's March 2026 survey of 1,120 U.S. adults, more than 93% of AI users took at least one more step before acting. 62% searched Google, 58% visited the business's website, and 52% clicked the sources the AI cited. Only about 7% acted on the recommendation with no further research.

Does being cited in AI answers still matter for SEO if people double-check?

Yes, but it's only part of the job. An AI citation puts you on the shortlist. The checking step decides whether people pick you. That means your Google results, your website, and your reviews need to back up what the AI said about you. If any of those don't match, the recommendation probably won't turn into a customer.