Brands are picking creators a machine will quote

What changed in the creator strategy brief#

Something moved in the creator strategy brief this year, and it did not arrive with a press release. Digiday reported on 4 September 2026 that brands now want creators who can win over humans and machines at once. At the performance agency Tinuiti, AI visibility was roughly a tenth of the factors weighed when building a creator strategy six to eight months ago. It is now closer to a quarter, and something like 60 per cent of the roster raises it unprompted.

Follow that trail slowly. The figures are interview-sourced. No survey, no sample frame, no stated methodology, and no published research linking creator selection to citation outcomes at all. One agency described its own internal weighting to a reporter, which tells you how briefs are being written rather than what works. Testimony from the field is worth hearing. Testimony is not evidence, and a creator strategy built on it alone is a guess with a chart on top.

To its credit, the reporting marks its own limit. Rate cards have not moved, and Digiday notes that the data “is setting the direction not the price”. Creators are folding citations into media kits to stand out rather than to charge more. So the theory has changed the brief without yet changing the budget, and that gap is the useful part: it is the window in which a marketing lead can think rather than spend.

Here is the awkward part. There is real research on what large language models cite, run at scale with a stated design, and it does not say what the AI visibility pitch says. Before re-scoring a roster on machine readership, ask what a citation actually buys, because a model can cite you and still describe you wrongly. That distinction decides whether a creator strategy buys an asset or an accident.

Those figures come from Cited but Not Verified, a benchmark of fourteen closed-source and open-source models whose opening premise is that “these citations cannot be reliably verified”. Link validity holds. Topical relevance holds. Factual accuracy does not. A creator can be quoted, credited and correctly linked inside an answer that gets the product, the price or the claim wrong, and the brand carries that error into a buying decision without ever seeing the sentence that caused it.

What an AI content strategy can actually change#

Set the vendor claims aside and read the best measurement anyone has published. What Gets Cited built a controlled test that injected exactly two candidate sources into a model context and recorded which one the first citation marker pointed at. Across six models the authors ran “252,000 trials, repeated paired comparisons under one factorial program over 18 content factors”, each pair differing in one factor, with source order counterbalanced to separate content effects from position bias.

The result is the sentence every AI content strategy deck should open with. Mixed-effects models showed that “topical relevance and list position are the biggest drivers of being cited first”. Explicit price information and a recent timestamp helped consistently. Completeness and trust cues added smaller gains. And then the clause that ought to reprice half the market: “formatting-only edits have little impact”.

Read that against what is being sold. Schema markup, question-shaped headings, bulleted answer blocks, the machine-readable rewrite of a page that already said the same thing: those are formatting edits, and the study puts them last. What sits first is being demonstrably about the thing the reader asked about, the oldest brief in the trade and the hardest to fake. Note too that the author discloses exercising the checklist in an early internal pilot at Sprinklr, so the paper is not disinterested either.

One page, two readers, and only one of them has eyes. That is the whole problem in a sentence, and it is far less exotic than the tooling suggests. A creator strategy that writes for a machine and a person separately produces two thin artefacts and pays twice. A creator strategy that writes one page a knowledgeable human would actually recommend tends to satisfy both, because relevance is precisely what the machine was measuring.

The platform documentation agrees, and it is free, official and routinely ignored. Google Search Central states that “There are no additional requirements to appear in AI Overviews or AI Mode”. It says site owners do not “need to create new machine readable files, AI text files, or markup”, and there is “no special schema.org structured data that you need to add”. The company running the web’s largest answer surface says the formatting layer is not the lever.

Its sister document is older and plainer. Google’s guidance on helpful, reliable, people-first content asks whether a page offers “original information, reporting, research, or analysis” and whether it is “written or reviewed by an expert or enthusiast who demonstrably knows the topic well”. Nothing on that scent trail is new. It is the same quarry the trade has chased since 2011, wearing a fresh acronym.

Cohley sells a brand-to-creator content platform, so that post advertises the theory as much as it describes it, which is exactly why it is useful here. Note what it claims. Creator content can influence how brands are surfaced: plausible, and unmeasured. It does not say by how much, in which categories, or against what else the same budget could buy. Set the vendor voice beside the trial data, and only one of them has a sample size.

None of this makes AI visibility fake. Models do cite, citations do carry brands, and the same study that demoted formatting found price information and recent timestamps helped consistently. Publishing prices and dating pages is actionable, cheap and boring. It is not the product most of the market is packaging, and it costs a fraction of an AI content strategy retainer.

The citation you win may still get you wrong#

Start with the click, where theory meets the ledger. Pew Research Center tracked 900 United States adults across 68,879 Google searches in March 2025 and reported that “Google users who encountered an AI summary also rarely clicked on a link in the summary itself”. Rarely is doing real work there: “This occurred in just 1% of all visits to pages with such a summary”.

Sit with that number. A citation inside an answer is seen far more often than it is followed, which makes it closer to a billboard than a click: a genuine branding asset and a poor performance one. Any creator strategy that swaps measurable creator-driven traffic for unmeasurable citation presence has moved money from a countable outcome to an inferred one, and the plan should say so plainly.

Pew found something else that ought to cool the panic. “The most frequently cited sources in both Google AI summaries and standard search results are Wikipedia, YouTube and Reddit”. The citation layer is no fresh meritocracy waiting for a clever brand to outfox it. It rewards the same durable destinations the open web already rewarded, which is why the shortcut through the thicket keeps leading back to the same path.

Then there is accuracy, which the industry keeps treating as somebody else’s problem. The same benchmark found fact-checking accuracy falling by roughly 42 per cent on average across two frontier models as tool calls scaled from two to 150, and its authors put the conclusion in a single clause: “more retrieval does not produce more accurate citations”. More machinery, more mistakes.

None of this is new. Search Engines in an AI Era, published in October 2024, ran a study with 21 participants comparing answer engines against traditional search and quantified “common limitations” including “frequent hallucination, inaccurate citation”. Twenty-one people is a small sample, so read it as background rather than proof. It still tracks: the reliability problem was on the record two years before this sales cycle began.

So the honest framing for a client sounds like this. A citation is a share-of-voice signal with a weak click, an uncertain accuracy rate and a host who says formatting will not get you in. Worth tracking. Not worth rebuilding a roster around, and certainly not worth a premium nobody has justified with numbers.

The other reader, and why brand visibility in AI search can backfire#

While marketers prowl after the machine, the humans have formed a view, and it is not flattering. WordPress VIP commissioned a survey in which “60% say AI in brand messaging is a turnoff”, and in which “86% say they always or sometimes explore it after receiving a summary”, so the source page still matters enormously. Its own Future of the Web write-up adds that “74% of consumers say the internet feels less human than it did 10 years ago”.

Methodology first, house rule. The release states the survey “was conducted by Talker Research and surveyed 2,000 respondents in April 2026”, split between 800 enterprise decision-makers and 1,200 United States adults. That is an opt-in panel, not a probability sample, so the percentages are directional. Read them as the shape of a mood, not a national measurement.

One more caution, since this release has been quoted carelessly all summer. It carries three separate 60 per cent figures: consumers calling AI in brand messaging a turnoff, enterprises reporting increased traffic from AI answer platforms, and the average share of enterprise reach now arriving through third-party platforms. Three questions, three bases. Anyone quoting one as another has not read the page.

That bottom bar is the sharpest. Sixty-one per cent were not sure or could not name a business using AI well in its messaging, and “another 16% said they do not believe any business uses AI well at all”. Two years of AI-forward brand work, almost no recalled winner. Chasing brand visibility in AI search while your published voice reads as machine-made loses you the second reader to reach the first.

The human audience, meanwhile, stays measurable and large. Edison Research at SSRS found that “29% of those age 13+ in the U.S. use TikTok daily”, that “64% have used TikTok to discover new music” and that 78 per cent of weekly users recall ever seeing or hearing an advert there. That study was “weighted to match the gender, age, and geography of the U.S. 13+ population”, which is what a properly sampled creator audience number looks like.

Writing a creator strategy for two readers at once#

The practical conclusion is unglamorous and cheap, which is roughly why nobody sells it. Brief for relevance and genuine subject authority, and one page serves both readers. The machine, when measured, selected on relevance. The human selects on whether the thing sounds like it came from somebody who knows. Same instruction, different coats.

That does not mean a content strategy agency has nothing to offer here, only that the honest version of the offer is different. A content strategy agency earning its retainer in 2026 is buying you subject authority, original reporting and a page a knowledgeable person would recommend, then measuring what that does to citations as a consequence rather than promising it as a deliverable up front.

For brand creator partnerships the change is smaller than the noise suggests. Keep hiring on category authority, audience fit and credible content, then add one question rather than a whole new scoring system: does this creator already turn up in machine answers for our category, and is what those answers say correct? Brand creator partnerships passing both are worth a premium. The rest are worth the old rate.

Two obligations do not soften because a machine is reading. The ASA influencers guide is written from a creator’s perspective but notes it “is also useful for brands and agencies”, and the FTC endorsement guides warn that a platform disclosure tool is not automatically enough: the regulator “would evaluate whether the use of the tool by itself clearly and conspicuously discloses the relevant connection”. A citation harvested from an undisclosed advert is a compliance problem in a growth costume.

One measurement discipline is worth adding, and it is modest. Track brand visibility in AI search as a brand metric, with a brand metric’s tolerances: sample a fixed set of category prompts monthly, record whether you appear, and record whether what is said about you is right. Nobody logs that second column, and it decides whether a citation is worth having.

Spend the rest where the arithmetic works. The same discipline runs through our SEO and GEO work, our content marketing practice, the brand strategy that decides what a brand is genuinely authoritative about, and the paid social carrying creator work to an audience. It is the argument folkfox made about unpriced creator deals and about AI Overview reporting: measure the measurable, and price the rest as a bet.

The pattern repeats across the week. A media review ending in consolidation, a dating app monetising the audience it already had, a casino brand selling direct and a blockchain network paid down one route tell one story: durable growth sat in the asset already owned, not the newest tactic with the best deck.

So keep the trail simple. Hire for authority, write for relevance, disclose the money, log the accuracy, and let citations arrive as a by-product of work that was worth citing. That is a creator strategy you can defend in a board meeting, and unlike the alternative it does not rest on a claim nobody has measured.

Frequently asked questions#