Distrust of AI content doubled in a year. What that means for your feed.
The useful finding of the last year is not that people dislike AI content. It is how fast they started to.
A 2026 Fractl study found the share of consumers saying heavy AI use would decrease their trust in a brand went from 20% in 2025 to 40% in 2026, with 14% saying it would decrease significantly. That is a doubling in twelve months, and it happened while the tools were getting better, not worse.
Which tells you the problem is not quality in the abstract. It is recognition.
"AI slop" is now a reputational category
The phrase has stopped being an aesthetic complaint. Analysis of public conversation shows mentions of AI slop skew heavily negative, with more than half expressing frustration or dislike.
That matters because the label is applied to your post by a reader in about a second, without appeal. Once it lands, everything downstream — the offer, the argument, the credibility — is discounted.
And the tell is rarely a factual error. It is the shape of the thing: the three-item list where two would do, the "in today's fast-paced world" opener, the paragraph that says something is important without saying anything about it.
The mechanism is authenticity, not automation
The most useful research here is a March 2026 systematic review in the American Impact Review covering 35 studies of consumer responses to AI-generated marketing content. Its finding is precise and worth internalising:
Perceived authenticity is the primary mechanism determining whether AI content earns trust or erodes it.
Not accuracy. Not polish. Whether it reads as though a person meant it.
This is good news, because it means AI is not the liability. Being indistinguishable from everyone else is. Sprout Social's 2026 data lands in the same place from the other direction: the single thing users most want from brands on social is human-led content.
The quiet failure: your voice flattens
The visible risk is publishing something obviously generated. The expensive one is slower.
When a brand leans hard on automation, its voice does not collapse. It converges. Sentence lengths regress to the mean. Opinions soften into positions nobody could disagree with. The specific, slightly awkward things that made the account recognisable get smoothed out, because smoothing is what these models do.
Six months later the account is perfectly competent and completely interchangeable, and nobody can point at the post where it went wrong — because none of them did. All of them did, a little.
Where AI genuinely helps
Use it where there is no voice to lose. That line is easy to apply:
Good uses. Resizing and reformatting. Generating twenty headline variants to react against. Turning a transcript into a rough outline. First-pass research and summarising. Alt text. Scheduling and the operational work around publishing. Checking whether you have already covered a subject.
Bad uses. The opening line. The point of view. Anything containing a claim about your own product. Replies to real people. Anything where being wrong is embarrassing rather than inconvenient.
The dividing line is not difficulty. It is whether a regular reader would recognise the result as coming from you.
A test that takes ten seconds
Before publishing, read the post and ask: could a competitor post this verbatim?
If yes, it is not your content. It is category content, and the audience has seen forty of them this week. Add the specific number, the thing that went wrong, the opinion your legal team is mildly nervous about. That is the part that cannot be generated, because it is the part only you know.
What to do about it operationally
- Write the first line yourself. Always. It is the highest-leverage sentence and the one that determines whether the rest is read.
- Keep a specific in every post. A number, a name, a date, a failure.
- Say something falsifiable. A position that could be wrong is a position worth reading.
- Do not disclose defensively, and do not hide. Nobody needs a label on a resized image. Nobody forgives a fabricated case study.
- Read a month of your own posts in one sitting, once a quarter. Voice drift is invisible day to day and obvious in bulk. This is the only reliable way I know to catch it.
Used carefully, AI takes the parts of publishing that were never the interesting bit and gives you back the time to write the sentence that is. Used carelessly, it makes you sound like everyone else at a moment when 40% of your audience has started actively watching for it.
Sources
- The Brand Leader — Authenticity, AI anxiety, and the future of social media in 2026
- Tiger Tracks — AI slop fatigue and the value of authenticity
- Media A La Carte — What AI slop is and why it is hurting brands
- Nurdd — How brands are beating AI slop with authentic content
Frequently asked questions
Do consumers dislike AI-generated marketing content?
They dislike content that reads as generated. A 2026 Fractl study found the share of consumers saying heavy AI use would decrease their trust in a brand doubled from 20% to 40% in a year. A systematic review of 35 studies found perceived authenticity, rather than the use of AI itself, is what determines the response.
What is AI slop?
Low-effort, repetitive content that is obviously machine-generated: polished on the surface and signalling nothing underneath. More than half of public mentions of the term express frustration or dislike, so it functions as a reputational judgement rather than a neutral description.
Should brands stop using AI for social media?
No. The evidence points at how it is used, not whether. Use it for work with no distinctive voice to lose, such as variants, resizing, scheduling and first-pass research, and keep a human on the opening line, the point of view and anything a regular reader would recognise as yours.