What does a category look like when 564 million social posts and fewer than 5,000 prompts disagree about it? As of November 10, 2026, the newest answer comes from Cision's State of Search and Social 2027, which compared 564.61 million posts with 4,678 search queries and AI prompts and found the two surfaces telling different stories about the same products, which is a problem for any team that still reports them as one number.
Two data surfaces that never meet
The figures come from the release of the report, which also notes that search volume for AI-related queries rose 2,400 percent over four years while social conversation carried data-centre concerns and pushback against low-quality AI output. Curiosity and approval are different signals, and a dashboard that adds them together will describe a market that does not exist.
Toyota 13.8 and 4,000 YouTube sources
The automotive example is the clearest illustration of the split. Toyota was named in 13.8 percent of brand mentions in the relevant sample of car-related answers, yet its own website accounted for only a fraction of the sources those answers cited, while YouTube supplied more than 4,000. A brand can dominate the sentence and still not control the evidence behind it.
Owned pages are the smallest slice
That has a practical consequence for review programmes. If the sources shaping an answer are creator videos, editorial reviews and forum threads, then the useful work is not another landing page but making sure the third-party descriptions of the product are accurate and specific. An analysis of the study makes the same point more carefully: the report is vendor-produced and the methodology sits behind a form, so it justifies a measurement audit rather than a causal claim that coverage produced the citation.
Magnesium agrees on nothing
The magnesium glycinate example is the one most review teams should read twice. Social discussion of the ingredient concentrated on stress relief, while the sampled AI answers gave more weight to bone support and cardiometabolic support. Neither surface is wrong, but a brand that only watches social will brief its content team on a benefit the answers are not repeating.
What a review brief should carry
A useful brief now needs three columns rather than one: what shoppers say, what the answers say, and which sources each one leans on. Where the three disagree, the disagreement is the finding. Where they agree, the claim is probably safe to repeat, and the brand can spend its effort on being the source rather than on being mentioned.
Where sentiment analysis stops working
Most consumer feedback programmes were built to score a review, not to explain one. A star average tells a category team whether shoppers were satisfied; it does not tell them which claim an assistant will repeat when a buyer asks what the product is for. That is a different job, and it needs a different unit of analysis, which is the sentence rather than the score.
| Surface | What it measures | What it cannot tell you |
|---|---|---|
| Social posts | Volume and emotional temperature of conversation | Whether the people posting are buyers |
| Star ratings | Satisfaction after delivery | Which benefit the shopper came for |
| Review text | Specific problems and outcomes | How often the theme appears across the category |
| AI answers | What an assistant says when asked to compare | Whether the citation reflects the brand or the publisher |
Read as a set, the four rows explain why sentiment analysis alone is no longer enough. A brand can hold a healthy rating and still lose the comparison question, because the answer is built from a handful of sources that repeat one benefit the brand never claimed. Consumer feedback has to be read at the level of the claim, not the mood.
Where the channel shift shows up
The surface split arrives while the channel mix is moving. Online food and beverage sales rose more than 15 percent in the four weeks to October 4 while physical store sales fell almost 3 percent, according to Circana data reported by Grocery Dive. More of the purchase decision is happening in a place where a summary, not a shelf, is the first impression.
That makes the review page a different asset than it was three years ago. It is no longer only a conversion tool for a shopper scrolling a product page; it is raw material for the paragraph an assistant writes about the category. Reviews that name the product, the problem and the outcome survive that compression, and reviews that describe a feeling do not.
A five-field answer audit
The audit is cheap because it reuses work the team already does. What changes is the discipline of keeping the question set fixed and the answers stored, which turns a monthly argument about visibility into a record that anyone can re-check. Without that record, every discussion of AI visibility ends the same way, with two teams quoting two different numbers and no way to tell which one is describing the customer.
FAQ
Is the Cision study a measurement of the market?
No. It is a vendor-produced study comparing social volume with sampled search and prompt activity. It is useful for spotting disagreement between surfaces, not for estimating category size.
Why does the Toyota example matter for reviews?
Because the brand was named far more often than its own pages were cited. The sources shaping the answer were creators and publishers, so earned description matters more than owned copy.
What is the fastest way to run this audit?
Fix 50 buyer questions per category, record the answers with dates and platforms, and log every cited domain. The discipline of a fixed question set is what makes the result comparable over time.
Does this only matter for consumer categories?
It matters wherever buyers ask comparison questions before purchase. A B2B software category behaves the same way, with analyst pages and community threads playing the role that YouTube plays in cars.
