What AI Content Costs You in Brand Trust.

Seven percent of consumers say that obviously AI-generated content in a brand’s marketing makes them trust that brand more.

Thirty-two percent say it makes them trust the brand less.

That is from Klaviyo’s 2026 AI Consumer Trends report, run with the research firm Datalily. Most social teams turn a number like that into an argument about whether to use AI at all, which is the wrong argument to be having.

AI is good at the parts of social nobody sees and weak at the part everybody sees. The work worth doing is drawing that line inside your workflow rather than inside your values statement, then naming who owns the last edit before anything ships.

The numbers are not close

Klaviyo surveyed 8,000 consumers aged 18 and over in December 2025 across the US, the UK, France, Germany, Spain, Italy, Australia and Singapore, with Datalily as its research partner. The findings were published in March 2026.

The question was specific: “What is your reaction to brands that clearly use AI-generated content in their marketing?”

• 7 percent said it makes them trust the brand more
• 32 percent said it makes them trust the brand less
• 61 percent said it makes no difference

Read the neutral majority honestly. Most people genuinely do not care.

But among the people who do react, it runs more than four to one against you, and the effect holds even in the friendliest group. Klaviyo’s own segment breakdown has 39 percent of self-described AI enthusiasts saying they would trust a brand less for it.

What the question actually measured

The word carrying the weight there is “clearly”.

Nobody in that survey was asked whether they minded a transcript being generated automatically, a first pass at a brief being drafted by a model, or footage being sorted by software before an editor opened it. They were asked about AI they could see in the finished thing.

That distinction gets flattened every time the figure is quoted, and it is where all the practical decisions live. The jobs that sit safely on the unseen side are covered in how AI should actually be used in social media content teams.

The research is not evidence that AI in production costs you trust. It is evidence that AI in the output does.

Budgets are moving the other way

Canva’s 2026 marketing and AI study, run with The Harris Poll, surveyed 1,415 marketing leaders at organisations of 500 people or more alongside 3,547 consumers across seven countries. Ninety-seven percent of those leaders use AI in daily creative work, and 99 percent said they plan to increase their AI investment.

The consumer half of the same study asked about advertising, and it pointed the other way. Seventy-eight percent said they prefer human-made ads. Seventy percent said they can usually spot an AI-generated one because it feels like it is “missing its soul”.

The gap is not really a disagreement about the technology. One side is buying speed. The other side is judging the output.

Visible is not entirely your call any more

Even when you say nothing, the platforms increasingly say it for you.

Meta announced in February 2024 that it labels images posted to Facebook, Instagram and Threads where it can detect industry-standard AI markers, meaning C2PA and IPTC metadata plus the invisible watermarks the major generation tools embed. It said at the time it could not yet detect those signals in video and audio from other companies, so that content relied on the uploader declaring it.

TikTok started reading C2PA Content Credentials in May 2024 and auto-labelling images and videos that arrive carrying that metadata, the first video platform to do it.

Detection is patchy and depends entirely on metadata surviving the trip, so the absence of a label proves nothing. What has gone is the option of quietly not mentioning it.

The two layers of every post

Every post has a layer the audience meets and a layer it never sees. Only one of them was in that survey.

The unseen layer is everything before the post exists: research, transcripts, tagging, asset naming, scheduling logic, a first pass at a brief, the reporting summary nobody enjoys writing.

The visible layer is narrower than most teams assume. The words in the caption. The voice and face in the video. The image itself. Every reply that goes out under your name.

Sorting between the two takes about ten seconds. Could a customer tell? If the answer is yes, or even maybe, a named person owns that output before it ships.

Where the cost actually gets charged

Trust is a soft word for a hard outcome, so it helps to look at the places it gets priced.

YouTube’s monetisation rules draw the same line in writing. To earn from a channel, content has to “be your original creation” and must “not be mass-produced, generic, repetitive, or manipulative”. In July 2025 the rule previously called “repetitious content” was renamed “inauthentic content”.

The policy does not ban AI. It expressly allows creative tools used “to assist in delivering a unique, well-researched, or creative narrative, like using AI to edit your video scripts or generate a unique background visual”. What it will not pay for is output assembled from a generic template with none of the creator’s own perspective in it.

A platform wrote that line into policy before most brands wrote it into process.

Put the line in the workflow, not in your principles

“We use AI responsibly” is a sentence, not a control. It lasts until the first Friday you are behind.

A control is a step somebody has to complete before a post moves. Something like:

1. Generated imagery of your product, your premises, your staff or your customers does not ship.
2. AI never speaks as the brand in public. No auto-drafted comments, no auto-drafted DMs.
3. Every caption gets a human rewrite of the first line and the last line, at minimum.
4. The voice, face and script in any talent-led video are human, and named on the brief.
5. If a piece could not survive being labelled, it does not go out unlabelled.

Rule five does most of the work. It turns a values question into a publishing decision, which is the only form a rule survives in.

Who owns the last edit

Most AI quality problems trace back to an ownership gap rather than a tool choice. Something got generated, was reviewed by nobody in particular, and went out because it was sitting in the queue.

Name a person per format, not per channel. Whoever owns the last edit on a Reel is not automatically the right person to own the last edit on a customer complaint, and pretending otherwise is how a reply goes out sounding like a support macro.

Then write it down where the work happens. Our social media approval process template is the short version of that: who drafts, who checks, who signs, and what each of them is actually looking for.

The tells your audience already reads

Your regular followers have seen a hundred of your posts. They know what you normally sound like, which makes them better detectors than any tool.

• Captions that describe the image instead of adding anything to it.
• An opening line that could sit on top of any post in your category.
• Replies that thank someone for their feedback without answering the question.
• Generated product shots where a detail is subtly wrong: the label, the stitching, a shopfront that is not your shopfront.
• Scripts with even pacing, no hesitation and no opinion in them.
• Everything arriving in neat threes, because that is the shape the models default to.

None of that is a detection problem. It is a craft problem, and it shows up as a slow decline in saves, shares and replies rather than as one bad post.

What good looks like

A team with AI everywhere upstream and none of it in the last mile.

Research is summarised in seconds. Transcripts are automatic. Long-form footage arrives pre-cut into candidate clips. Reporting comes with a first draft of the narrative attached. Briefs reach a reviewable state the same day they are asked for.

Then a person writes the caption, records the voice, picks the frame and answers the comment.

The team ships more than its headcount suggests it should, and the work still sounds like somebody from your brand, because it is.

How NBK thinks about AI in a social team

We treat this as a workflow question rather than a technology question. The brands that get into trouble are rarely the ones using the most AI. They are the ones with no named owner on the visible layer, usually because nothing else in the process had a named owner either.

That is the work behind workflow and ops consulting: fewer stages, clearer ownership, and a definition of done that includes who touched the words last.

Next step

If your output has started to feel faster but flatter, the constraint is almost always in the process rather than the tooling. NBK can help find it and rebuild the workflow behind the content.

Until then, there is one question worth asking at your next content review. On this post, who owned the last edit?