AI slop is not bad content. It is expected content.

Two agency owners recorded a ten-minute podcast about AI slop, and the most useful sentence in it is a definition. In the Sure Oak conversation, Mac from Campfire (a storytelling agency) says slop is "derivative content": a model writes from what has already been said, so you never say anything new.

Note what the definition leaves out. It does not say slop is bad. It says slop is expected. That is a much more uncomfortable problem, because "bad" you can fix with an editor.

The ad everybody recognises in two seconds

Nick, who runs an SEO and GEO agency for enterprise brands, describes an ad from a competitor of one of his furniture-manufacturer clients: a good-looking chair, and a half-dressed AI woman who walks in and does a thing. His verdict was that everybody knows what it is. Nobody is offended and nobody remembers it.

Mac's version for writing is blunter. If a brand is a reputation, and you say the same things everyone else says, "then you really don't have a brand at all". The blog is there, the content engine hums, and the brand quietly disappears. As an example, they point to HubSpot, which had a period when it stopped getting credit for content that repeated what books and periodicals had already said.

I work as a strategist, and this matches what I see when I read drafts. The sentences are fine. The structure is fine. You could swap the logo for a competitor's and nobody would notice, and that is the whole problem.

Why the temptation is so strong

Scaling a content team with AI is cheap and fast, and the output looks like work. The speakers do not argue against using it. Both of them use AI in almost everything they do, and Mac says the "check the results" disclaimer is not just a legal line, because the model really does make mistakes. They argue about what you hand over. Process is fine. Voice is not.

There is also a second audience that cares. Nick says there is growing pushback against AI-written content from human readers and from AI systems themselves. That is his view from the GEO side, not a published study, so treat it as a field report. But it fits the logic: a system that summarises the internet has no use for one more page that summarises the internet.

Six things to follow

• Take risks. Say things that have tension, even counterculture. Safe content is the easiest thing for a model to produce, which makes it the least valuable thing for you to publish.

• Invest in a real brand. Brand language, a logo and wordmark pack that still works on a hat or a mug, colors, typefaces and line art, built by a strong designer. Those are the building blocks you feed AI afterwards.

• Make it personal. Tell stories about yourself, your employees and your clients.

• Show the human. The most underrated part of a blog post is who wrote it. A reader (and an LLM) clicks the author name and asks why this person has the authority to say it with such confidence.

• Speak from where your authority is. Twenty years in a business means unique insight. Those insights should be the points that make you stand out.

• Publish original data. Only your brand can own research it ran itself. In a world where everything is derivative, unless you actually do research or learn something new from your customers, you are rearranging other people's sentences.

The About Us page gets a bonus mention as the most underrated page on a site. It is unique content, and it is where they expect LLMs to look to work out who you are.

My take: slop is a positioning failure, not a writing failure

I would push the argument one step further. When every competitor in a category uses the same tools with the same prompts, their output converges. Copy that sounds like the category average is what you get when you ask a model for "a blog post about X" without bringing anything of your own. The model is not lazy. You gave it nothing to be specific about.

So the fix is upstream of the writing. Before you generate anything, you need the thing a model cannot know: a position you would defend in a pub, a customer observation nobody has published, a number from your own data. We built Selfstorming around this idea. The tools make you work from sourced cases and structured techniques first, so the AI is expanding your thinking rather than replacing it.

A fair caveat on the source: it is an opinion podcast. There are no statistics in it, and I would not quote any of it as data. What it offers is a clear definition and a practical checklist from two people who use AI daily, and the checklist survives contact with reality.

One thing to do this week

Open your three most recent posts. For each one, answer two questions: what in here could only have come from us, and who is the named human that stands behind it? If the answer to the first is "nothing", delete a paragraph of general advice and replace it with one number from your own customers. If the answer to the second is "nobody", put a real author on it to rewirite it.

That is an hour of work, and it is the part no model will do for you.