On May 24, The Guardian published a piece that reads like a marketing post-mortem written before the funeral. Aisha Down quoted an anonymous central London PR account director with a clean number: roughly half of the pitches she receives overstate the AI involvement of the product they describe. Half. Not the long tail. The median pitch.
That is what marketing’s first real governance incident looks like.
The number on its own would be a curiosity. What makes it operational is what surrounds it. The same week, Standard Chartered’s CEO Bill Winters publicly apologized for calling workers displaced by AI “lower-value human capital” during a Bloomberg interview on May 22. Allbirds, the sneaker company, pivoted its public narrative in April toward “acquiring AI GPUs,” a sentence that means nothing in the context of selling shoes and means something specific to investors. PR practitioners interviewed by The Guardian cited press releases for “AI-powered basketball hoops” and “AI-powered lasers” as the working examples of where the genre has gone.
These are not outliers. They are the shape of the year.
The diagnosis: marketing has no equivalent of code review
Engineering shipped its governance layer in the last 18 months. Pre-commit hooks. CI pipelines. Static analyzers. Eval suites for LLM features. A human-readable code-review step before anything reaches main. The work is unglamorous, the controls are imperfect, and they exist. When an engineer writes is_ai_powered = True on a function that calls a regex, four other engineers see the line before it ships.
When a PR firm writes “AI-powered” in a press release about a basketball hoop, the equivalent review does not happen. The agency drafts it. The brand approves the spirit. Legal scans for libel. Nobody asks: is this factually true? What does the model do? Where is it called? What is the underlying mechanism? The closest analogue to a code review in PR is a pull-quote review, and pull-quote reviews check tone, not truth.
Half the pitches overstate AI involvement because the function that produces those pitches has no formal mechanism to catch the overstatement. The PR director quoted by The Guardian did not describe a malicious industry. She described a default. When the incentive is to get coverage and the verification step does not exist, the median output drifts toward the most coverage-friendly framing, which right now means “AI-powered.”
This is what an unguarded surface looks like at scale. We named the pattern in Your Marketing Team Just Became a Governance Team and again in Marketing’s Governance Reckoning. The Guardian reporting is the field evidence.
The Standard Chartered moment
Standard Chartered matters in a different way. Bill Winters is a chief executive of a global bank. The phrase “lower-value human capital” did not appear in a press release a junior wrote at 11 p.m. It appeared in a live Bloomberg interview. He apologized within 48 hours. The apology is the data point: the company recognized, fast, that the framing was a brand-safety event.
Read the sequence carefully. A CEO speaks. The market processes. The CEO retracts. That is the unmediated path between executive language and reputational consequence. It is also the path that marketing, in most large firms, is now structurally responsible for governing, because the alternative is letting the CEO improvise on live television with no review of the language patterns the company has decided are off-limits.
This is the same point made by Allbirds in reverse. A consumer brand publicly explaining its pivot toward “AI GPUs” is producing language that is both factually thin and obviously aimed at the investor who has lost patience with sneakers and is willing to forgive losses if the word AI is present. The market reads that signal correctly: not as conviction, but as positioning. The damage is to long-term credibility, which is the asset marketing exists to protect.
The external clock: the SEC is already enforcing
The voluntary window for fixing this is closing. The U.S. Securities and Exchange Commission has been bringing enforcement actions for AI-washing in securities filings since 2024. In March 2024 the SEC settled with two investment advisers for $400,000 over false AI claims. In June 2024 it charged Joonko’s founder with defrauding investors of $21 million by claiming proprietary AI matching that did not exist. The Office of the Investor Advocate flagged AI-washing as a 2025 priority. The legal mechanism is operational, the case law is being built, and it applies to any communication that touches a securities filing, which in a public company is most external communication.
What the SEC is doing for filings, plaintiffs’ attorneys will do for consumer claims, and regulators in the UK and EU will do under their own frameworks. The European Commission’s AI Act creates disclosure requirements that already conflict with the casual “AI-powered” claim. The UK’s CMA has signaled scrutiny of AI marketing claims under existing consumer-protection law. The pattern is the same one engineering saw with security and accessibility a decade ago: voluntary discipline before mandatory enforcement, then mandatory enforcement for everyone who did not adopt voluntary discipline.
The firms that install factual-claim review now will look, in 24 months, like the firms that adopted SOC 2 before customers required it.
What the review layer actually looks like
A factual-claim review for marketing copy is not exotic. It is four questions, asked before any external surface ships, by someone with the authority to say no.
First: is there a model in this product? Yes or no. Not “machine learning informs,” not “AI-enabled,” not “powered by.” Is there a model that runs on input and produces output. If no, the word AI does not appear in the copy.
Second: if yes, what does the model do? One sentence in plain language. “It ranks candidates by skills match.” “It generates draft email replies.” “It classifies invoices.” If the answer is more than one sentence, the copy needs more specificity, not less.
Third: what is the user-visible effect? Speed, accuracy, coverage, cost. A number, a range, or a comparison. “Reduces classification time from 15 minutes to 30 seconds.” If you cannot produce a measurable user-visible effect, the AI is not the story.
Fourth: who signs off that the previous three answers are true? Name. Not role. Not team. Person.
That is the entire mechanism. It is a 15-minute review. It is also the difference between half your pitches overstating AI involvement and none of them doing it.
The counter-argument, acknowledged
Some marketing leaders will read this and say the function has always managed factual accuracy. Legal reviews exist. Compliance reviews exist. The industry has been writing about AI claims for two years.
This is true and not enough. Legal reviews check for actionable false statements, not for the soft inflation that produces half the pitches overstating AI. Compliance reviews check disclosure requirements, which in most jurisdictions still do not specifically cover AI claims for non-regulated products. The two-year conversation produced essays and panel discussions, not pre-publication checklists with named owners. The Guardian’s reporting documents a function that has been aware of the problem and has not built the mechanism to fix it.
The same way engineering’s awareness of security in 2010 did not produce SOC 2 by accident. Someone had to build the checklist, name the owner, run the audits, and accept that some campaigns would be slower and some claims would be smaller. The firms that did that work first now sell to enterprise customers without a six-month security review every time. The firms that did not are still doing the work, just under deadline.
Do this now
If you lead marketing or communications, three actions before Friday.
Pull every external claim about AI your firm has made in the last 90 days. Press releases, product pages, sales decks, executive speeches. List them. The list itself is the audit.
For each claim, run the four questions above. Mark each line green, yellow, or red. Yellow means the claim is defensible but vague and should be tightened. Red means the claim is wrong and needs correction or retraction.
Name the owner of the factual-claim review going forward. One person. Calendar a weekly 30-minute review block. Put it in the comms calendar with the legal review and the brand review. Make the review a published gate that copy must pass before external release.
The function exists in engineering and works. The function does not exist in marketing and the bill is arriving. The PR director quoted anonymously by The Guardian was describing an industry waiting for permission to do the work. The permission is the calendar block.
This analysis synthesizes AI Washing: PR Firms Scrambling to Rebrand (The Guardian, May 2026).
Victorino Group helps marketing and communications leaders install factual-claim review processes before regulators do. Let’s talk.
All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →
If this resonates, let's talk
We help companies implement AI without losing control.
Schedule a Conversation