The Truth about Trust part 3: AI now writes the content, and decides who sees it

AI now sits on both ends of the exchange between B2B buyers and the people trying to reach them. It is writing more of what gets published, and filtering more of what gets seen. Trust has kept pace with neither.

Ask a B2B marketer whether they trust AI-generated content and most will hedge. Ask whether they used AI to write something this week, and the hedging stops: two-thirds now say they use AI tools frequently or constantly to create their own content.

Ask a business leader the same question from the buying side, whether they trust an AI chatbot’s answer, and most say no. Ask whether they’ve used one to help source or shortlist a vendor recommendation, and the answer flips hard: 82% say yes, according to G2’s 2026 buyer behaviour research. That number is climbing too.

Neither habit is matched by confidence in the result. TrustRadius’s 2026 survey of B2B buyers found 47% now trust online resources less than they did a year ago, and the sceptical share is growing, not shrinking. Asked how much they trust what AI specifically tells them, 60% say only sometimes, 20% say very often, and just 2% say always. Most of the growth in AI usage is sitting on a foundation of active scepticism, not confidence.

That is the shape of the problem on each side. More is being made with AI, more is being filtered by AI, and trust in either has not caught up.

How much B2B content is now generated

Graphite, a research group that tracks web publishing at scale, sampled 55,400 URLs from a full web archive and ran them through three separate AI-detection models – see the report here. Their finding for the first quarter of 2026: 49.9% of newly published articles were primarily AI-generated, against 50.1% human-written, having briefly tipped past the halfway mark the previous quarter. That is near enough to parity that it no longer reads as an edge case. On the production side, the same shift shows up in how B2B marketers describe their own work: two-thirds now say they use AI tools frequently or constantly to create content.

None of that is damning by itself. Plenty of AI-assisted writing is careful and well-sourced, and plenty of human-written material never was. The problem is what happens at the volume this is now running: with roughly half of what gets published on a given day AI-generated and climbing, and no reliable way for a reader to tell which piece was written with care, credibility becomes a coin flip rather than a judgement call.

The homogenisation risk in AI-written content

This is where it would be easy to overreach, so it is worth being precise about what the evidence actually says. Research comparing outputs from 22 large language models against 102 human writers, covered recently in Nature, found that AI-generated responses clustered far more tightly around each other than the human responses did, even though individual AI outputs were often rated more original than individual human ones. Separate research tracking scientific writing found articles grew measurably more similar in style and content after generative AI tools became widely available. A third strand found something stranger still: people who had used AI assistance on a task gave noticeably more uniform answers weeks later, on a different task, with no AI involved at all. Researchers are calling it a creative scar.

Phil Treagus-Evans, CEO of Giraffe Social Media, made the same point from the practitioner’s side in a recent episode of the Trust & Influence in B2B podcast. The distinction he cares about isn’t AI against human, it’s supervised against unsupervised, using the tool to sharpen a thought against using it to skip having one. That is exactly the condition the research above is describing too. None of this is a verdict on any individual writer’s judgement. The problem sits one level up, in what a system trained to predict likely, plausible, well-formed continuations is structurally built to reward: the median, the already-said, the version of an idea most compatible with everything that came before it. That is a condition on what a lot of AI-assisted production drifts toward, not a verdict on the people or the tools doing the producing.

How B2B buyers are using LLMs to research and shortlist vendors

On the other side of the same exchange, buyers are asking AI to do more of the filtering for them. G2’s 2026 buyer behaviour research, drawn from more than 1,000 B2B software buyers, found 82% had sourced a software recommendation from an AI chatbot in the past two years. Forrester’s own buyer research goes further: business buyers now cite generative AI and conversational search as a more meaningful source of information than any other channel, twice as often as they did the year before. The 2026 Reuters Institute Digital News Report, drawn from roughly 2,000 respondents in each of 48 countries, finds weekly use of AI chatbots for news and information has climbed from 7% to 10% globally in a single year.

None of that usage looks like confidence. The same Reuters Institute research puts trust in AI chatbot answers at 20% globally, the lowest score of any source it tracks. Rajat Bhatnagar, SVP of Growth at HG Insights, summed up what TrustRadius’s 2026 survey of B2B buyers found: “Buyers are using AI to move faster, not to think less. They want speed, but they still want the efficiency of AI synthesis and the confidence of verified sources, and those two things have to coexist.” The same survey found the mechanism behind that: 94% of buyers who use AI fact-check its responses at least some of the time. They say this, and use the tools anyway, because reading everything themselves stopped being realistic a while ago, and the load they are managing rarely gives them a better option in the moment. That is a rational trade under time pressure, not naivety.

The zero-click reality of AI-mediated search

The mechanics matter here, because delegation is not the same as summarising. AI summaries are no longer an edge case at the point of search: Similarweb’s tracking puts them on roughly 43% of US Google searches by mid-2026, up from about 15% a year earlier, other trackers land at different points using different methods, but all show the same steep climb. Pew Research Center tracked 68,879 real Google searches from 900 US users over a month in 2025. When an AI summary appeared above the results, users clicked through to a normal search result 8% of the time, against 15% when no summary appeared, a rate cut almost in half. Only 1% clicked a link inside the summary itself. More strikingly, more than a quarter of those sessions ended there. No further click, no further reading, just the summary and then nothing.

Sarah Aird-Mash, CMO at Future Group and formerly head of marketing at Google EMEA, was already writing about this in mid-2025, before it had been measured this precisely, using the industry term zero-click to describe it. When I interviewed her for the Trust & Influence in B2B podcast, she pointed out something worth carrying forward: AI engines do cite their sources, but the citation lands as an afterthought rather than a reason to click through. The specific adoption figures she cited then have already gone stale, this is a space where a year-old snapshot doesn’t hold, but the shape of the behaviour she was describing has since been measured properly, and it holds up. A filter that frequently ends the search before the buyer reaches anything at all is not surfacing the best of what exists. It is deciding, on the buyer’s behalf, that they have their answer.

Why AI content creation and AI search are the same problem

Put the two halves of this together and they stop looking like two separate trends. The tool a buyer uses to filter a flood of content is, increasingly, sorting through content that was produced by the same kind of tool. That is not a coincidence, and it is not a neutral fact. A filter is only as good as the variety it has to work with, and the material it is drawing on is itself converging toward the median, for exactly the reasons covered above.

Provision and consumption are not two problems sitting next to each other. They are the same mechanism, viewed from either end of the exchange, and each one is quietly making the other’s job harder: more homogeneous material going in, a filter with less genuine distinction left to find, and a buyer at the other end who never sees what got smoothed out along the way.

Why the gap between LLM usage and buyer trust matters

Set that against what actually moves a purchase decision, and the tension sharpens. TrustRadius’s 2026 research found buyers rank product demos, free trials, their own prior experience, and user reviews as the most influential resources when choosing a vendor, the concrete, checkable, first-hand kind of evidence. Vendor-produced marketing content ranked last. Worth a raised eyebrow here: TrustRadius is a reviews platform, and its own research found that the thing buyers trust most is reviews, which is also the thing TrustRadius sells.

None of this is a mystery to the brands choosing AI content anyway. It wins internally on speed and cost, a brief that used to take a week now takes an afternoon, and that calculation is real. What it overlooks is that the trade barely touches trust. Faster and cheaper was never the same as more trusted, and the best-written piece of content in the world doesn’t help much if it never reaches a buyer who’s delegated the reaching to a system that doesn’t trust it either.

That leaves two separate problems dressed up as one. Making sure your material is actually legible to the systems now doing the filtering is a real, practical response, and worth taking seriously on its own terms.

It does nothing to answer the harder half. If content itself is no longer where trust gets built, something else has to do that work instead, and right now nobody, including the brands spending the most on AI content, has settled what that is.

This is the third piece in a series drawn from eighteen months of interviews and research into how B2B buyers actually decide who to trust.

The Truth about Trust part 1: From centralised to atomised – why B2B trust has a new owner

The Truth about Trust part 2: More noise isn’t more trust