What we check: ten kinds of AI-generated content mistakes
AI drafting tools are fast and fluent, and their failure mode is that they are confidently wrong in ways that read fine. An AI content quality check is a human editor reading for those failures on purpose. Here are the ten things our editors look for in every article, with an example of the kind of note you get for each. The examples are fictional. Companies like “Northwind Plumbing” and “Lumen Analytics” do not exist.
1. Facts and claims
Wrong, outdated, unverifiable, or invented facts: statistics, quotes, names, dates, prices, product features, and anything legal, medical, or financial. The editor checks what can be checked and flags what cannot. A claim with no source is a problem even if it happens to be true.
AI wrote: “Tankless water heaters last 30 years, three times longer than a standard tank.”
The note the editor sends: Remove “30 years” and “three times longer” unless you can link to a manufacturer or utility source that says so. We could not find one. Keep the sentence qualitative (“typically outlast a standard tank”) or cite a specific warranty.
2. Hallucinated specifics
Fake studies, fake sources, tools that do not exist, misattributed quotes, and invented examples presented as real. These are the most damaging mistakes because they are the most convincing. A named study with a year and a percentage reads as authoritative until someone looks for it.
AI wrote: “A 2023 report from the Digital Commerce Institute found that stores with live chat convert 42% more visitors.”
The note the editor sends: We could not find this organization or this report. Delete the sentence. If you have a real study on live chat and conversion, cite it with a link; otherwise make the point without a number.
3. Logic and structure
Contradictions, circular arguments, a conclusion the body does not support, missing steps in a how-to, repetition, padding, and sections that exist to make the article longer. AI drafts produce the shape of an argument. The editor checks whether the argument is there.
AI wrote: Section two says “the best approach is to track every metric you can.” Section five says “track only the three metrics that matter and ignore the rest.”
The note the editor sends: These contradict each other. Pick one. Section five is the stronger position, so cut the sentence in section two and the two paragraphs that follow it.
4. Reader fit
Does the page deliver what the title promised, and is it written for the person who will actually read it? AI drafts explain basics to experts or skip basics for beginners, because they have no idea who is reading.
AI wrote: Under the title “Advanced retargeting for ecommerce stores,” the first 400 words explain what a cookie is.
The note the editor sends: Delete the first three paragraphs. Someone searching for advanced retargeting knows what a cookie is. Start at the paragraph beginning “Segmenting by cart value.”
5. Voice and tone
The generic AI voice: hedging, tidy triads, “in today’s fast-paced world,” “it’s important to note,” over-signposting, empty transitions, and sentences that say nothing. Also tone that wanders between formal and chatty within one piece. We flag the specific sentences that do no work, not the article for sounding like AI.
AI wrote: “In today’s fast-paced digital landscape, it’s important to note that choosing the right software can be a complex decision.”
The note the editor sends: Delete this sentence. It is the first line of the article and it says nothing. Open with the next sentence, which makes the actual point.
6. Specificity
Vague where it should be concrete: no real examples, no numbers, no decisions, no opinion. The editor marks every place where a reader would ask “like what?” or “how much?”
AI wrote: “Our onboarding process is fast, flexible, and designed with your team in mind.”
The note the editor sends: Replace with what actually happens: how many days onboarding takes, the steps, and who from your side needs to be involved. If it is one call and a shared checklist, say that.
7. Accuracy for your business
Claims about your own product or service that are wrong or risky: features you do not have, promises you cannot keep, guarantees your terms do not back, and compliance-sensitive statements in regulated fields. The editor reads the rest of your site to check the article against it.
AI wrote: “Northwind Plumbing guarantees same-day service anywhere in Illinois.”
The note the editor sends: Your service-area page lists three counties and your booking page says “next available.” Change this to match, or this paragraph is a promise you will be held to.
8. Mechanics
Grammar, usage, terminology that changes from paragraph to paragraph, headings that do not match the content under them, broken lists, and formatting that makes the page harder to read. The smallest errors, the fastest to fix, and the ones readers notice first.
AI wrote: A heading reads “Three ways to lower your bill” and is followed by four items.
The note the editor sends: Change the heading to “Four ways,” or cut the fourth item, which repeats the second.
9. Search intent
Checked lightly, not as an SEO audit. Would the page satisfy someone who searched for its topic? Is it missing the thing people actually want, or obviously stuffed with a keyword? If the title promises a comparison and the body never names a product, the reader leaves.
AI wrote: An article titled “Best accounting software for freelancers” discusses “the importance of choosing the right tool” for 900 words and never names one.
The note the editor sends: Name the three tools you would actually recommend, say who each is for, and give one reason to pick each. That is the whole reason someone opened this page.
10. Originality and risk
Passages that look lifted or near-duplicate, trademark and brand misuse, and images or captions that mislead, where we can see them. We do not run plagiarism software. An experienced editor recognizes a definition that appears word-for-word on a hundred sites, and a stock photo captioned as your team.
AI wrote: A caption under a stock photo of five people at a whiteboard reads “The Lumen Analytics team planning our 2026 roadmap.”
The note the editor sends: This is a stock image. Change the caption to describe the topic, or replace the photo with one of your actual team.
What we don’t flag
Style preferences that are just taste. If you use the Oxford comma, contractions, sentence fragments, or one-line paragraphs, and you use them consistently, that is your voice and we leave it alone. Formatting you chose deliberately, like bulleted summaries, bold key phrases, or short sections for mobile readers. Opinions we happen to disagree with, as long as the article supports them. Your pricing, positioning, and product decisions, unless the article states them wrongly.
What you get is one email with the problems numbered, quoted, and written as instructions your AI can apply. See how it works, or send us a page.