Machines Read More of It. People Trust Less of It
Two curves crossed this week without anyone announcing it: the share of your traffic that is a machine, and the share of your audience that holds machine-made words against you.
What people actually say about machine-made words#
Sixty per cent of surveyed consumers call AI in a brand's messaging a turnoff, not a feature. That is a pricing signal, not an opinion.
Ask a marketing team whether ai content marketing works and you will get a productivity answer: faster drafts, more variants, lower cost per asset. Ask the audience and you get something else entirely, and the gap between those two answers is where the next two years of ai content marketing budgets will be decided.
In its 2026 survey of 1,200 US consumers, WordPress VIP reports that 60 per cent say AI in a brand's messaging is a turnoff, not a feature, and that 74 per cent say the internet feels less human than it did ten years ago. Those are perceptions rather than purchase records, and they should be read as such. They are still the clearest statement yet that ai generated content carries a cost the efficiency case never books.
Why ai content marketing splits its own buyers#
The more surprising number is on the buy side. In a companion survey of 800 enterprise decision-makers, WordPress VIP reports that 85 per cent believe publishing AI-generated content without human review erodes brand trust, and 91 per cent say it is important that their content adopts a more human tone. Seventy two per cent also say publishing speed and freshness still matter for AI-driven discovery.
Read those three together and the tension is explicit rather than hidden. The same leaders who want a human tone want machine speed, and most ai content strategy documents resolve that by pretending the trade-off does not exist. It does. The honest question for ai content marketing is not whether to use the tools, it is which parts of the trail you are willing to let a machine walk.
Neither survey publishes a fielding date or a full methodology on the pages we read, so treat the exact percentages as directional. The direction is not ambiguous, and a fox reads the wind rather than waiting for the weather report.
What it looks like when provenance fails#
The risk to ai content marketing is not theoretical, and this week it had names attached. On 3 September Press Gazette reported that it had identified four tech writers whose identities could not be verified, all of whom had published articles promoting cryptocurrency and blockchain projects. Publications including Hacker Noon and Information Age removed the pieces after being contacted.
Press Gazette states plainly that it asked publications who had published work by the four to verify their credentials, and that none were able to do so. It follows an earlier investigation into four other finance writers who between them had more than 1,000 published articles. Note the careful framing there, and keep it: unverifiable is not the same as fabricated, and the reporting does not claim to have established who or what produced the copy.
The audience is not waiting for a policy to form a view. A thread on Reddit this morning captured the split more honestly than most research does, with one creative describing an argument at home about whether it matters how something was made if the result looks good.
My partner doesn't care if AI art or other product was made with AI as long as it "looks good" and he enjoys it.
That is the whole commercial question in one sentence. A meaningful slice of any audience genuinely does not mind. Another slice minds intensely, and tends to be the slice that writes reviews, tells friends and remembers. Ai content marketing that assumes the first group is everybody is making a bet it has never actually priced.
Meanwhile, the audience that is not human#
While ai content marketing teams argue about whether readers can tell, the composition of the traffic itself has quietly changed. On 3 September Press Gazette reported analysis by 51Degrees covering three billion website visits in the year to 29 May 2026, across sites in telecoms, advertising, media, retail, financial services and technology.
The individual figures are worth stating precisely. AmazonBot recorded 6.43 million sessions, more than 4 per cent of all web visits over the period. Meta-External-Agent recorded 4.23 million, almost 3 per cent. AhrefsBot took 2.17 million and BingBot 1.62 million. ClaudeBot recorded roughly 288,000 sessions and the OpenAI search crawler roughly 129,000.
Together the two largest accounted for 63 per cent of AI bot sessions, and the four leading crawlers made up 85 per cent of the crawl sessions generated by the 72 bots identified. That concentration matters more than the totals: a content programme optimising for every crawler is optimising for a long tail that barely visits, chasing scent through undergrowth where nothing lives.
Bot traffic on those sites grew 62.5 per cent over the twelve months, reaching 13 per cent of web sessions in May 2026 against 8 per cent a year earlier. So roughly one visit in eight is a machine, and the machines are consolidating around a handful of very large readers. Any ai content marketing plan written before that shift needs revisiting.
What a content programme does about it#
The instinctive response in ai content marketing is to pick a side, either banning the tools or leaning in and hoping nobody notices. Both are lazy, and the week offers a better model from an unlikely place.
Press Gazette reported on 2 September that Politico has moved from more than six backend platforms to a single headless content system, and that the change let it pull together more than 20,000 rulings on detention practices and analyse them with AI assistance. The machine did the volume work on a corpus no human could read. Journalists still did the judgement, and that is the ai content marketing template worth stealing.
That is the division of labour worth copying. Use the tools where the task is scale and the output is an input, not a publication. Keep human authorship where the task is judgement, claim-making and reputation. Most failed ai content marketing work gets this exactly backwards, automating the byline and hand-crafting the spreadsheet, and most ai content strategy documents never notice.
Does ai marketing actually work in practice?#
Ai content marketing works where it is measured against the right thing. Faster production is easy to demonstrate and largely beside the point if the output erodes the trust the brand was buying. The harder measurement, and the one worth building, is whether audiences behave differently towards content they believe a person made.
There is a sharper commercial reason to get this right now. Publishing economics are tightening around exactly the places brands rent attention. Press Gazette reported that Guardian Jobs is closing, winding down from 4 September, with the publisher citing UK vacancies at their lowest level in over five years, at 707,000 for May to July. When a publisher closes a revenue line, the surviving inventory gets more expensive and more crowded, which raises the stakes on every ai content marketing placement you buy.
Write down what may be machine-drafted, what must be human-authored, and what always carries a named byline. Publish it if you can. Undeclared practice is the thing that becomes a story later.
If you place bylines on third-party sites, check those sites can verify their own contributors. Press Gazette found four whose publications could not.
Move budget towards named experts, original data and creators the audience believes, since that is what stays scarce as words get cheap.
One visit in eight is a crawler, concentrated in four of them. Make the pages legible and fast without writing for the crawler instead of the reader.
Add a provenance question to brand tracking. If 60 per cent of consumers say AI messaging puts them off, that belongs on a dashboard rather than in a footnote.
None of this is a rejection of the tools. It is an argument for ai content marketing that knows which of its two audiences it is writing for at any given moment, because they now want opposite things, and only one of them buys anything. Our content marketing practice starts every engagement with that split, and our search and answer-engine work treats the crawler as a distribution channel rather than a reader to flatter.
The fox that survives a hard winter is not the one that ate fastest. It is the one that knew which trail led to the den. Ai content marketing rewards the same instinct: produce less, stand behind more, and let the machines have the parts nobody was ever going to read anyway. Anyone weighing that up against headcount should look at what this actually costs before assuming automation is the cheaper path.
Frequently asked questions#
does ai marketing actually work?
It works for scale tasks measured against the right outcome. Drafting speed is easy to prove and largely irrelevant if trust falls. WordPress VIP reports that 60 per cent of 1,200 surveyed US consumers call AI in brand messaging a turnoff, and that 85 per cent of 800 surveyed enterprise leaders think publishing without human review erodes brand trust. Measure trust, not throughput.
How much of my website traffic is actually AI crawlers?
On the sites 51Degrees monitored, AI bots reached 13 per cent of web sessions in May 2026, up from 8 per cent a year earlier, across three billion visits in the year to 29 May 2026. Your own share will differ by sector, but roughly one visit in eight being a machine is a reasonable planning assumption.
Which AI crawlers actually matter?
Very few. AmazonBot and Meta-External-Agent together accounted for 63 per cent of AI bot sessions in the 51Degrees analysis, and the four leading crawlers made up 85 per cent of sessions from 72 bots identified. Optimising for the long tail of crawlers is effort spent on visitors who barely arrive.
Should I disclose that content was AI-assisted?
A written internal rule is the minimum, and public disclosure is increasingly defensible. Undeclared practice is what turns into a story later, as the Press Gazette investigation into unverifiable bylines shows. Decide what may be machine-drafted, what must be human-authored, and what always carries a named byline.
Why are creator costs rising if AI can produce content cheaply?
Because provenance is the scarce input, not words. Digiday reported that usage and exclusivity are additional costs in creator deals almost 100 per cent of the time. As machine production makes copy abundant, value migrates to the named person an audience believes, which is exactly what cannot be generated.
What is the first thing to change in an ai content strategy?
Split the work by task type rather than by tool. Use automation where the job is volume and the output is an input, as Politico did in analysing more than 20,000 rulings. Keep human authorship where the job is judgement, claim-making and reputation, which is everything that carries your name.
Read more on this topic#
The jug did not get fuller. The pouring got narrower
Where web traffic actually went, and who caught it.
Read the pieceBrands are picking creators a machine will quote
Creator selection when the reader you are optimising for is a model.
Read the pieceNobody publishes what the last deal closed at. So everyone guesses
Why creator pricing stays opaque, and what it costs you.
Read the pieceThree files tell you the whole strategy
What your robots file admits about your machine-readership policy.
Read the pieceTwo audiences. Opposite preferences. One budget.
folkfox builds content programmes that hold up with readers and with the crawlers, without pretending those are the same job. If your output has gone up while your trust signals have not, that is a conversation worth having.
Want folkfox in your Google results and AI answers? Set folkfox as a preferred source.