Deciding what deserves to be published is now the scarce skill in AI content
In a June 2026 release, Gartner reported that 49% of 307 US consumers surveyed in March said generative AI has made the quality of available content worse. Among Gen Z and millennials, 57% agreed. The sample is small and US-only, so the figure is directional for a UK brand rather than a benchmark, but the direction is clear enough to plan against.
That consumer verdict matters because the production side has moved in the opposite direction. Teams under pressure to innovate reach for tools that produce more, faster. The pitch for most AI content software is throughput: more drafts, more variants, more posts per week. What those tools do not supply, and what no procurement checklist seems to measure, is the judgment to decide which of those drafts should ever leave the desk.
That is the scarce skill now. The willingness to withhold, rather than the ability to generate.
Why volume became the default metric
Innovation pressure inside marketing functions tends to be measured in outputs: cadence, reach, number of pieces shipped. When a leader is asked to show how the team is innovating, a chart of rising publication volume is easier to produce than a chart of rising reader trust. AI tools slot neatly into that reporting habit, because they move the volume number quickly and visibly.
The problem is that the same move is available to every competitor at the same cost. Kate Muhl, VP analyst at Gartner, puts it plainly in the release: AI-generated content is increasing the volume of media consumers encounter but not necessarily the value, which raises the stakes for brands to be recognisable, credible and intentional.
Read that alongside two other numbers from Gartner's work this year. A separate February 2026 survey of 328 US consumers found 17% rely on AI summaries and 16% use AI chatbots to search for products or services. And 59% prefer to do several media activities at once. Attention is fragmented, routed through summarisers, and allocated by readers who have already concluded that most of what they see is worse than it used to be. In that setting, the marginal AI-drafted post does not add reach. It adds to the pile the reader is trying to filter out.
What the mechanism actually is
The behavioural pattern underneath is familiar. Under time pressure, people defer to the fastest available option and underweight the slower costs. For a brand, the slower cost is reputational: a feed of adequate, interchangeable posts quietly teaches the audience, and the systems summarising for them, that this publisher is not worth quoting.
I think the mistake is treating AI content tools as production machinery when the harder question is editorial. A production machine should run at capacity. An editorial function should refuse most of what crosses its desk. The two logics do not coexist well, and most teams have swapped the second for the first because the first is easier to defend in a quarterly review. A withheld draft looks like wasted effort on a dashboard. A shipped one looks like work.
What editorial judgment looks like in practice
A useful starting move is to separate two decisions that AI tools tend to collapse into one. The first is whether a piece can be produced. With current tools, the answer is almost always yes. The second is whether it should be published under the brand's name. That decision needs a human who can say no without having to justify the saved effort.
A few questions make the second decision easier to hold:
- Does this piece say something the brand's audience could not read, in substantially the same form, on three competitor sites this week?
- Is there a specific reader whose problem it answers, named concretely enough that the writer could picture them?
- Would the author be willing to be quoted on the strongest sentence, by name, in a trade publication?
Pieces that fail all three are the ones that quietly push the 49% figure further in the wrong direction. They are also, inconveniently, the pieces that AI drafting tools produce most fluently, because fluency is what they are optimised for. The judgment that catches them is not a prompt. It is a person with the authority to end a draft without rewriting it.
Where this points next
If consumer perception of AI-produced content continues on its current line, the differentiation available to brands will come from editorial restraint rather than editorial output. The teams that benefit will be the ones that can show, internally, that their publication decisions are deliberate: that something was written, considered, and then not shipped.
That is harder to report on than volume. It is also the number that will matter when AI summarisers are deciding which publishers to lean on and which to treat as background. A practical forward move is to track withheld pieces alongside shipped ones, and let the ratio describe what the function actually does. A team that ships everything it writes is not editing. A team that ships half of what it writes has started to.
Source: Gartner, "Gartner Survey Finds 49% of U.S. Consumers Say GenAI Has Made Content Quality Worse", 9 June 2026.