Shoppers read the AI answer, then click to check it: Fractl's quiet click data for CPG brands
A new Fractl and Search Engine Land study pairs a consumer survey with clickstream data. The clicks that survive an AI answer are slower, more deliberate and mostly about checking. We charted the parts that matter for packaged goods.
Most AI search studies count clicks that disappeared. A new study from the content agency Fractl, published with Search Engine Land on November 2, looks at the clicks that are left and asks why people make them. Its answer is short: people read the AI summary first, then click to check it.
The study has three parts. A survey of 1,007 US adults about how much they trust AI summaries and what they do next. A Datos clickstream panel of roughly 11,100 research and comparison keywords tracked from 2024 to 2026. And a query audit across 28,327 queries and about 862.6 million monthly searches in eight verticals. Fractl's own caveat is worth repeating up front: the clickstream tables weight every keyword equally, so they describe direction, not population-wide click-through rates.
We pulled out the figures that matter most for a brand that sells packaged goods, and charted them from the article's numbers.
What people do when an AI answer appears
The survey asked a simple question: when an AI answer shows up at the top of the results, what do you usually do?
Show the data
| Behavior | Value |
|---|---|
| Read it and click a link to double-check | 56% |
| Scroll past it to the regular links | 22% |
| Read it and stop there | 16% |
| Search again with different words | 3% |
| Don't notice AI summaries | 3% |
Source: Fractl, The Quiet Click (n = 1,007 US adults). Self-reported behavior. Figures as published by Search Engine Land.
Trust has grown, but it has limits. 64% of respondents said they trust AI summaries at least somewhat, and 45% said they trust them more than a year ago. At the same time, 77% said they have caught an AI summary being wrong at least once.
The stakes decide how far people go. For a quick, low-stakes fact, 55% said the AI summary is enough. When comparing options, 48% said they need to click out. Before a final decision or purchase, 71% want to see a real source.
Shopping is lower stakes than health, until it is not
Fractl also asked which topics most need a real source. The order is a useful reminder that most grocery and household purchases feel low risk to the person making them.
Show the data
| Topic | Value |
|---|---|
| Health and wellness | 72% |
| Personal finance and banking | 53% |
| Education | 27% |
| Shopping | 27% |
Source: Fractl, The Quiet Click. Survey of 1,007 US adults.
For a CPG brand, that cuts two ways. A shopper choosing paper towels may be happy with whatever the AI summary says. A shopper choosing a supplement, baby food, an allergen-free snack or a sunscreen is in the health column, where 72% want a source. Many CPG categories sit closer to health than to shopping in a buyer's head. Those are the categories where the verification click is most likely, and where your own page and your reviews need to be ready for it.
The click is slower and more deliberate
The clickstream panel shows the shape of the remaining clicks changing between 2024 and 2026.
| Measure | 2024 | 2026 |
|---|---|---|
| Median time from search to click | 2.5 seconds | 6.0 seconds |
| Zero-click bounces | 43% | 34% |
| Continued searching (re-queries) | 54% | 63% |
| Click-through rate | 3% | 3% |
| Clicks with reads of 60 seconds or more | 9% | 10% |
| Visits under 10 seconds | 71% | 67% |
Source: Fractl and Datos clickstream panel, as published by Search Engine Land. Equal-weighted per keyword; directional.
Fewer people walk away with nothing, more keep searching, and the click itself comes later. That fits the survey: people read, decide whether the summary is enough, and click when they want proof. It also fits what other studies show about volume. Ahrefs found a 58% lower click-through rate for the top-ranking page when an AI Overview is present. Fewer clicks, then, but the ones left are doing a specific job.
The page under the answer is usually a list
The query audit looked at which page formats win the organic slots directly beneath an AI answer on research and comparison queries. One format dominated.
Show the data
| Format | Value |
|---|---|
| Listicles | 81% |
| Resource pages | 11% |
| Forums and Reddit | 5% |
Source: Fractl query audit (28,327 queries, about 862.6M monthly searches). Remaining share not broken out in the published article.
Consideration-stage queries made up 71% of the search volume Fractl analyzed. That is where a shopper compares brands, and it is mostly owned by "best of" pages and roundups written by someone else. Fractl's earlier work across more than 8,500 brands found that the roughly 4% of brands that outperformed their SEO footprint in AI search kept showing up in third-party roundups, expert lists and comparison reviews.
Where the escaping clicks go
When users left the search engine they started on, they did not mostly land on publishers.
Show the data
| Destination | Value |
|---|---|
| Google's other surfaces | 35% |
| Amazon | 13% |
| 7% | |
| YouTube | 5% |
| Bing | 3% |
Source: Fractl and Datos clickstream panel. About 40% of all exits were hops between search engines. NerdWallet, Forbes and PCMag each drew about 1%.
Amazon at 13% is the number we would circle. A shopper who reads an AI answer and wants to check it often goes to the Amazon listing, which lines up with PYMNTS data showing 59% of recent AI-assisted purchases were completed on Amazon. When people did land on content pages, product pages took 31% of those clicks and reviews 20%.
What people trust, and why they click
Ranking and trust turned out to be different things. Forums and community threads topped the survey's trust ranking at 21%, followed by news articles at 18%, AI search summaries at 17% and official company websites at 17%. Forums were also the most clicked format, at 22%.
The reasons people gave for clicking after an AI answer point the same way:
- A trusted expert or professional: 24%
- Original data or research: 23%
- Real customer experiences or reviews: 17%
- A Reddit or forum thread with real people: 15%
Nearly three quarters of respondents (73%) said they add Reddit or another forum to a search at least sometimes. Gen Z does it most: 55% add a forum often or every time, against 41% of millennials, 25% of Gen X and 11% of baby boomers. Trust in the AI summary itself runs the other way by age. 81% of boomers expressed at least some trust in AI search summaries, compared with 58% of Gen Z.
For a brand, the order matters. A company's own website ties with the AI summary in the trust ranking, behind forums and news. That does not make the brand site useless. It means the site is one witness among several, and the shopper is weighing it against reviews, forum threads and the press. A claim that appears only on your own page carries less weight than the same claim repeated by a reviewer, a dietitian quoted in a news story or a customer on Reddit. The practical response is not to post on forums at scale. It is to make sure the product facts other people repeat are correct, easy to find and the same everywhere, and to give experts and reviewers something worth quoting.
What we would do with this
Write for the verification click. If a shopper arrives to check what an AI told them, the page they land on should confirm or correct it fast. Put size, count, key ingredients, certifications and price where they can be seen without scrolling, and keep them identical to what your retailer listings say. Our tip on matching product facts everywhere covers the basics.
Get into the lists. With listicles holding 81% of the slots under AI answers, the most direct way to be present at the comparison stage is to be included on the roundups that already rank. That is digital PR and product seeding, not a new page on your own site. Our tip on earning the sources AI cites has the steps, and how assistants pick brands explains why third-party pages carry so much weight.
Treat the Amazon listing as a fact-check page. If 13% of escaping clicks land on Amazon, the listing is often where a shopper confirms the AI's claim. Reviews, Q&A and the bullet points need to answer the questions an assistant raises. Our Alexa for Shopping playbook is a good checklist for that.
Measure the quality of the click. Fractl's own advice is to look past click volume to engagement and downstream actions. For a brand, that means comparing conversion and time on page for visits from AI referrals, search and direct, not just counting sessions. Some teams track what assistants say about their products with tools such as SearchPanel from XENA Intelligence; a monthly spreadsheet of the same ten prompts works too.
The useful shift here is in how to think about the click you still get. It is not the start of the journey any more. It is the shopper checking whether the AI was right, and the page they choose to check against is the one that decides it.
Questions
Do people still click after reading an AI answer?
Many do. In Fractl's survey of 1,007 US adults published November 2, 2026, 56% said they read the AI answer and click a link to double-check it, 22% scroll past it to the regular links, and 16% read it and stop.
What kind of page ranks under an AI answer?
On the research and comparison queries Fractl audited, listicles held 81% of the winning organic slots and resource pages another 11%. Forums and Reddit accounted for 5%.
Sources
- The quiet click: Why people read AI answers and click anyway. Search Engine Land (Fractl research), Nov 2, 2026.
- Marketing research. Fractl.
- Update: AI Overviews Reduce Clicks by 58%. Ahrefs.
- Amazon Wins 59% of AI-Assisted Purchases as Checkout Becomes the Prize. PYMNTS, Oct 30, 2026.
How we work: every figure in this piece comes from the sources above, which we opened and read before publishing. Anything labeled illustrative is an example, not data.
Related tips
- How do you get into the sources AI answers cite?
- Should you publish comparison pages?
- Do reviews change what assistants recommend?
Keep reading
A randomized test of Google AI Mode found 18.8 points fewer clicks. What CPG brands should take from it
Most of what we know about AI search and clicks comes from correlations. This study randomly assigned real Google users to AI Mode, normal search or search with no AI, then watched what they did for a week. The results are stark, and the caveats matter.
Named is not recommended: two new studies on where AI brand picks come from
Most AI visibility reports count how often a brand is named. Two October studies looked at what the answer does with the name, and at which pages each engine reads before it decides. The engines turn out to recommend in very different ways, and the pages that move a recommendation are not the same on every engine.
Being famous isn't the same as being recommended by AI
New data from 2,720 assistant answers splits brand awareness from AI recommendation, and the two barely line up. For legacy CPG brands that's a warning. For challengers it's an opening.