Original insights by Si Quan Ong
30-second rundown
Key learnings:
- AI slop begins when responsibility disappears: Ahrefs uses AI in every article, but requires a human to understand the topic, choose the argument, verify the evidence, and own the published result. The Ahrefs blog generates an estimated 212K organic visits a month.
- Editorial gates expose hidden AI decisions: Separating the idea, outline, evidence, and draft lets a writer reject weak assumptions before polished prose turns them into an expensive-looking corpse.
- Saved time should buy stronger content: Use AI efficiency to add original research, useful tools, fresher data, and better presentation instead of merely flooding the web with more average pages.
This week: Choose one planned article and write a one-page brief naming the reader, promise, point of view, required evidence, and information unavailable in current search results. Ask AI to challenge the angle and identify evidence gaps, then let a named owner approve or reject the premise before any drafting begins.
Coming 3 months: Build an AI-assisted editorial workflow that pauses every article at the idea, outline, evidence, and draft stages. Require one accountable writer and a second knowledgeable reviewer to approve each gate, then review monthly whether published work contains original evidence, stays current, and produces qualified visits or enquiries.
Si Quan Ong has reported from inside the content factory with a fact that should make the anti-AI priests choke on their ceremonial keyboards: Ahrefs uses AI in every article. AI involvement does not create slop. Abandoned human judgement does.
Ong defines slop as content published without enough understanding, judgement, evidence, or original contribution to deserve attention. The creator avoids the hard thinking, then dumps the cost onto the reader, who must sort truth from padded landfill.
1. The Human Work Happens Before the Draft
AI made prose cheap, but it did not make good editorial decisions cheap. Before asking a model to draft, Ahrefs decides who the reader is, what the article promises, what point of view it will defend, which claims need proof, and what new information it can contribute.
The premise and evidence must exist before the prose. AI can map search results, challenge an angle, surface counterarguments, and identify missing support. It should not decide what deserves publication while everyone else stands behind the curtain pretending the slot machine is an editor.
The model also needs material unavailable from every competitor’s internet. Ong lists interviews, internal knowledge, proprietary data, demonstrations, failed experiments, and examples from real work. Feed it public consensus and it will return public consensus wearing a freshly ironed shirt.
Ahrefs supports this with a searchable source of truth containing facts, explanations, product details, and how-to guidance. Ong also dictates rough thinking because polished notes can remove the uncertainty and half-formed connections where useful ideas hide.
A running-shoe retailer could record how three models performed after six months of wet-weather use, including the failures. AI can organise the observations and help structure the comparison, but it cannot honestly invent the wear test, the soaked socks, or the judgement earned from doing the work.
2. Put a Human Decision at Every Gate
One-shot prompts conceal dozens of choices inside fluent prose. Research, angle, structure, claims, examples, and tone arrive fused into one confident package, so weak decisions look settled before anyone has inspected them.
Ahrefs breaks the process into separate outputs for research, content gaps, outline, and draft. A writer can inspect each stage, repair the instruction or evidence that caused a problem, and restart from the last acceptable point instead of repainting a condemned building.
Every stage needs permission to proceed. The idea gate asks whether the team has anything useful to add. The outline gate checks whether each section serves the promise. The evidence gate tests the important claims. The draft gate hunts for invented certainty, filler, and examples nobody earned.
This does not require an elaborate pipeline. A small team can approve the angle, build the outline, verify the evidence, and only then request prose. At each stop, the owner must be willing to demand research, change the argument, or kill the article.
The danger is AI-generated sunk cost. Six minutes can produce 2,000 polished words, and suddenly everyone wants to rescue them instead of asking why they exist. AI turns the writer into an editor before the writer has finished being a thinker, a neat little trapdoor disguised as productivity.
3. Spend the Savings on Work Readers Can Feel
The obvious efficiency play is to make the same article for less money and spend the savings on volume. Ahrefs does automate tedious work, including a data-refresh system that saves at least a day each month while updating twelve datasets more regularly.
But volume is only one return. Use saved time to make each article harder to copy. AI can help a writer analyse data, prototype a tool, improve presentation, or automate research that previously needed scarce specialist time.
That capability still needs restraint. High-stakes analysis may require a data scientist, developer, or subject expert. The point is not to dress every blog post in blinking machinery. It is to attempt valuable work that budget or capacity previously kept outside the fence.
Ownership closes the circuit. One knowledgeable person must understand the topic, validate the claims, explain the evidence, and attach their name to the result. Ahrefs also requires a second human with enough authority to challenge the premise, identify generic sections, and say no rather than rubber-stamp a robot’s homework.
Ong notes that heavily AI-generated pages can rank, including in top-three positions, but Ahrefs’ research found search performance tended to decline as AI use increased. That does not prove Google punishes AI. The likelier explanation is that companies use cheap execution to skip original work and manufacture more average pages.
The robots can research, interrogate, organise, draft, and refresh the files. Let them. But keep a human at the evidence locker and another at the exit, because content becomes slop the moment nobody understands, verifies, or stands behind what escaped into the desert.
