Merriam-Webster picked "slop" as its 2025 word of the year, defining it as low-quality digital content that AI produces in bulk. In April 2025, Ahrefs found that 74.2% of 900,000 newly published web pages contained some AI-generated content. So AI is already in the majority of new web pages. The real question is how you use it.

AI-assisted vs AI-generated content sounds like a debate about tools. It isn't. It's about how much human judgement goes in before you hit publish.

2 businesses can use the same model and get very different results. The first publishes a useful article with a clear point of view. The second publishes another page that sounds like every other page.

There's also a middle ground that plenty of guides skip. We call it AI-driven content: AI writes the first draft, and a person edits it properly. This article explains all 3 modes, shows where AI slop comes from, and gives you a 5-question test to check which one you're really doing.

AI-assisted vs AI-generated content: the 3 modes at a glance

The difference is who does the thinking: with AI-assisted content, a person writes and AI supports. With AI-driven content, AI drafts and a person edits properly. With AI-generated content, AI writes and nobody checks. The tool stays the same in all 3, and only the human judgement changes.

People use these terms loosely. Some say AI-led, others say AI-automated. Here's how we use them in this article:

• AI-assisted: You lead. You write the piece, and AI helps with research, structure, edits and tough questions. The ideas and the words are yours.
• AI-driven: AI leads the first draft. A person then edits it properly: adds first-hand experience, checks every fact, cuts the filler and takes a position.
• AI-generated: AI leads and nobody checks. A prompt goes in, a page comes out, and it's published with little or no human input. This is where AI slop comes from.

Think of it as a dial. At the human end, a person does the thinking and AI supports. At the other end, nobody does the thinking at all. Many teams work somewhere in the middle, which is why it's worth treating that middle as its own thing.

AI-assisted content: you write, AI helps

AI-assisted content is writing where the human is the author and AI is the support act. You bring the idea, the experience and the opinion. AI helps you get there faster.

In practice, that means using AI to research a topic, suggest an outline, tighten a rough paragraph, or ask the questions a sceptical reader would ask. The words on the page started with you. AI-assisted content creation is usually the slowest of the 3 modes, and that's the trade-off.

Say you're writing about why customers leave. You might use AI to organise your notes, suggest an order for your points and flag anything unclear. You write the argument yourself, using what you've seen across your own accounts. AI never decides what you think.

When AI-assisted is the right call

Use it when the piece depends on what only you know. Opinion pieces, founder posts, case studies and anything with a strong point of view all fit here. If a reader would pick it up because of who wrote it, don't hand the writing to a machine.

The risk is small, but it's there. If AI suggests a statistic or a claim, check it before it goes in, even when you wrote the rest yourself.

AI-driven content: AI drafts, a human genuinely edits

AI-driven content, or human-edited AI content, is the middle ground. It works when the edit has substance. AI does the heavy lifting on the first draft, and a person does the work that makes it worth reading.

Quality starts before the draft does. The draft is only as good as what you feed in, so give AI your notes, your sources and a clear point of view rather than a 1 line prompt. Ask it for something to react to, not something to publish.

The word "genuinely" matters here. A quick skim and a spell check isn't an edit. A genuine edit changes the piece, and it includes:

• Adding first-hand experience: a project you ran, a mistake you made, a number from your own work.
• Checking every fact, statistic and source against the original, not against what the draft claims.
• Cutting lines that any competitor could have published.
• Taking a position, so the piece says something rather than summarising everything.
• Rewriting in your own voice until it sounds like a person at your business wrote it.

Here's what that looks like on the page. Say an AI draft opens with a line about businesses increasingly adopting new tools. A genuine edit replaces it with something specific: a client problem you've solved, a number from your own reporting, or a decision you'd make differently now. Same draft, completely different piece.

There's a fair counterargument. Content platform Clearscope argues that a piece written by a subject matter expert, with AI helping, beats an edited AI draft every time. For original thinking and strong opinion, we agree.

AI-driven content suits explainers, guides and comparisons, where the value is clear structure and accurate information. It's a weaker fit when the whole point is your own perspective.

How we run it at Untapped

We use this approach for our own blog. Each post starts with research into current sources, then an outline that a person approves before any drafting begins. The draft then goes through separate checks for SEO, for phrases that sound machine-written, and for accuracy against the original sources. A person reviews it before it goes live, and this article went through the same process.

It isn't fast in the way pressing "generate" is fast. Each stage catches different problems, like a statistic that doesn't quite say what the draft claims, or a paragraph that sounds like everyone else. That catching is the point.

Our AI policy explains how we approach AI more broadly. If you're looking at how AI could fit into your own team's work, take a look at our AI engineering and products services.

AI-generated content: where AI slop comes from

AI-generated content is AI output published with little or no human input. A prompt goes in, a page comes out, and nobody asks whether it's accurate, useful or different from what's already on the topic.

That's where AI slop comes from. As the Associated Press reported, Merriam-Webster made "slop" its 2025 word of the year, and its definition covers low-quality digital content produced in quantity by AI. People now have a word for the thing they keep scrolling past.

What AI slop looks like

You can often spot it. The writing is vague and neutral, it avoids taking a side, and it could be published on a competitor's site with the logo swapped. It repeats common advice without adding anything from hands-on experience. Facts rarely get checked, so errors slip through.

What Google does about it

Google doesn't ban AI content. Search Engine Journal notes that Google has said since 2023 that it rewards quality however a page was made. When Google announced its scaled content abuse policy in March 2024, it said the policy applies whether content comes from automation, people or a mix. The target is publishing many pages mainly to manipulate rankings, not the use of AI itself.

The ranking data adds some nuance. Ahrefs looked at about 150,000 pages from the top 10 results for 100,000 searches in June 2026. It found that 54.7% of top 3 results scored under 20% on its AI detector, and 82.2% scored under 50%. Pages its detector scored at 100% AI made up 5.3% of top 3 results, so they can and do rank.

Even so, higher AI scores lined up with lower positions and fewer indexed pages. Ahrefs found signs of indexing for 49.28% of the least AI-heavy pages, against 40.35% of the most AI-heavy. Its authors call the gap meaningful but not disqualifying.

Treat this as a trend, not a rule. Ahrefs sells the detector behind these scores, and the scores only estimate how much of a page reads as machine-written. Search Engine Journal points out that the samples didn't measure depth, originality or accuracy. A high score is a reason to look at a page, not proof that it's weak.

What it costs you with readers

Readers notice too. A Gartner survey of 1,006 US consumers (May and June 2026) found that 65% think brands produce too much AI-generated content. It also found that 57% say this content has made them trust brand messaging less.

YouGov data reported by MarTech points the same way. Across the markets surveyed, 32% of consumers would trust a brand less if they knew its content was AI-generated, against 15% who would trust it more.

Gartner is also clear that people aren't rejecting AI outright. They're becoming more selective about what they trust. That's your opening: use AI, but make sure what comes out is worth trusting.

What this looks like in software development

The same 3 modes apply to code, and the gap between them matters even more, because bad code fails in ways bad copy doesn't.

• AI-assisted: A developer writes the code and uses AI as a pair. It suggests approaches, explains unfamiliar code and spots bugs. The developer designs the solution and owns every line.
• AI-driven: AI writes a first version and a developer reviews it carefully. That means reading every change, running the tests, checking security and making sure it fits the wider system.
• AI-generated: A prompt produces an app, and it ships with little or no review. This is what people mean by vibe-coded software.

The failure points differ from a blog post. An unchecked article might repeat a wrong statistic. Unchecked code can leave a secuity flaw open, struggle when live users arrive, or become hard for anyone else to maintain.

The test is similar. Could a developer on your team explain how it works? Has someone tried it with live data and actual users? Would you trust it with customer information?

AI-driven development suits well-understood work such as boilerplate, tests and standard features. For the parts that carry the most risk, like authentication, payments and your core logic, we'd lean towards AI-assisted. And if you've inherited something vibe-coded, a human review is the quickest way to find out what you actually have. We offer low-cost, fixed-price technical audits for vibe-coded products.  Get in touch here if that sounds like it would be of benefit to you.

How to tell which one you're doing: a 5-question test

Knowing the difference between AI-assisted and AI-generated content is easy. Knowing which one you're doing is harder. You don't need a detection tool to find out. Ask yourself these 5 questions about any piece before you publish it.

1. Could a competitor publish this unchanged? If you could swap the logo and nobody would notice, it's generic.
2. Does it contain something only your business knows? A project you delivered, a figure from your own work, a lesson from a mistake.
3. Did a person check every fact and number? Against the original source, not against the draft.
4. Does it take a position? If it only summarises, it adds nothing to what's already out there.
5. Would you put your name to it? If you'd hesitate, your readers will too.

Each no points to a specific fix. If question 2 is a no, ask the person closest to the work for 1 concrete example and add it. If question 3 is a no, trace every statistic back to its original source and cut any you can't verify. If question 4 is a no, decide what you actually think and say it in the first 2 paragraphs.

Mostly yes? You're working AI-assisted or AI-driven, and the process is doing its job. A few no answers mean the edit isn't doing enough, so add the missing experience, checks or point of view before you publish.

Mostly no? You're in AI-generated territory, and that's the moment to slow down. The fix is rarely to use less AI. It's to put more of your own judgement into the process.

Keep the judgement, use the tools

AI-assisted vs AI-generated content comes down to 1 question: who's doing the thinking? AI-assisted keeps you in charge. AI-driven works when a person edits with first-hand experience and checks every fact. AI-generated, where nobody checks, is how you end up with slop.

The tool matters far less than the process around it. Run your next piece through the 5-question test and see where it lands.

If you'd like a second pair of eyes on how your team uses AI for content, get in touch. We'll tell you straight where it's working and where it isn't.

FAQs

Does Google penalise AI-generated content?

Google doesn't ban AI content, and has said since 2023 that it rewards quality however a page is made. Its scaled content abuse policy targets publishing many pages mainly to manipulate rankings, whether AI or people produce them. Ahrefs data from 2026 shows heavily AI-flagged pages tend to sit a little lower and be indexed less often, though some still rank in the top 3.

Is AI-assisted content better than AI-generated content?

In most cases, yes. With AI-assisted content you supply the thinking, the experience and the opinion, while AI-generated content skips that step and tends to read as generic. Both use the same tools, so the difference comes down to human input, not technology.

What is AI slop?

AI slop is low-quality digital content produced in bulk by AI, usually with little or no human oversight. Merriam-Webster named "slop" its 2025 word of the year. The AI isn't the problem. The lack of checking, originality and value is.

Can people tell when content is AI-written?

Sometimes. Generic, neutral writing that never takes a side is the easiest to spot, and unchecked facts give it away fast. What matters more is how readers react when they know: YouGov data reported by MarTech found 32% would trust a brand less if they knew its content was AI-generated.

Should you tell readers you've used AI?

It's a judgement call, and the data cuts both ways. Some readers trust a brand less when they learn AI was used, but Gartner found people are becoming more selective rather than rejecting AI outright. Whatever you decide, make sure a named person stands behind what you publish.

What is scaled content abuse?

It's Google's name for publishing many pages mainly to manipulate search rankings instead of helping people. Google introduced the policy in March 2024 and says it applies whether the content comes from automation, people or a combination. The test is whether the pages help users, not how they were made.