Six things go wrong, and each one has a public record

Five checks that prevent all six

1. Open each source and check each fact, quote and number yourself.
2. Put one fact in the piece that a competitor could not have.
3. Agree with the client, in writing, when AI use is disclosed and to whom.
4. Use the agency's approved account for client material.
5. Do not publish pages in bulk to win search traffic.

1. It states wrong facts and invents sources

• Journalists at 22 public broadcasters checked more than 3,000 answers from four AI assistants. 45% had at least one significant issue, and 31% had serious sourcing problems (BBC and European Broadcasting Union, 22 October 2025).
• A syndicated summer supplement in the Chicago Sun-Times recommended 15 books. Ten did not exist (Chicago Sun-Times, 29 May 2025).
• Press Gazette named more than 50 apparently fake "experts" who appeared more than 1,000 times in the British press through PR and SEO pitches. Many had AI-generated copy (Press Gazette, 9 January 2026).

More cases: fabricated citations in an audit of ten models (Naser, preprint, 2026), and 18 non-existent legal citations out of 45 in one case in a joined High Court judgment (June 2025).

2. Everyone's work starts to look the same

• Writers who were given AI ideas wrote stories that judges rated higher. The stories were also more alike (Doshi and Hauser, Science Advances, 2024, 293 writers).
• In one brainstorming task, 94% of the ideas from people who used ChatGPT shared concepts with each other. Nine people gave their toy the same name (Wharton summary of Meincke, Nave and Terwiesch, May 2025).

Ad tools have the same cause. LinkedIn's Davang Shah gave it, as PPC Land reports: the tools sit on the same language models and draw on the same data, so outputs drift toward a mean.

3. Readers trust it less when they know or suspect

• A label of "AI-generated" lowered the perceived accuracy of a headline, including headlines that were true and written by people (Altay and Gilardi, PNAS Nexus, 2024, 4,976 people in the US and the UK).
• 89% of UK adults say it is important that content states it was made by AI. 68% say undisclosed AI use would lower their trust in a brand (YouGov, 12 August 2026).
• US adults rated content they believed to be AI-made 48% less trustworthy, whether or not AI made it (Raptive, vendor survey, August 2025, 3,000 adults).

The public asks for a label, and the label lowers trust. What works has the experiments and the one condition under which the penalty went away.

4. The advertiser stays responsible, and you may not own the text

The UK position, from CAP on 11 June 2026:

"If an AI tool produces something problematic, the advertiser remains responsible for any output used in their ads."

• There is no general UK duty to label AI in ads, and a label does not cure a misleading claim (ASA, 29 May 2025).
• Fake reviews are banned in the UK in any form (CMA guidance, April 2025). The US rule names AI-generated fake reviews (FTC, August 2024).
• In the US a prompt does not make you the author: "prompts alone do not provide sufficient human control" (US Copyright Office, January 2025).
• UK law still protects computer-generated works. The Government proposes to remove that protection and keep it for AI-assisted work (UK Government report, March 2026).

5. Client information goes into personal accounts

• 31% of UK workers who use generative AI at work do so without their employer's knowledge. 17% pay for a tool themselves. About half have had no formal training on safe use (The Next Web on a Deloitte and Ipsos survey, 16 September 2026).
• 27% of marketers pay for their AI tools entirely out of their own pocket, and another 26% share the cost with their employer (Social Media Examiner, July 2026, 681 marketers).

Clients have started to ask. Keri Bruce of the law firm Reed Smith told Digiday (10 September 2026): "Ultimately, people want disclosure. They want to know what tools are being used".

6. Search punishes volume that adds nothing

Google's spam policy covers "Using generative AI tools or other similar tools to generate many pages without adding value for users" (Google spam policies, updated 28 August 2026).

Its guidance on generative AI content (page last updated 1 October 2026) says:

"It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing."

Google's human raters give the lowest rating to pages where almost all the content is AI-generated or paraphrased "with little to no effort, little to no originality, and little to no added value" (Search Engine Land, 9 April 2025).