AI Content Quality Audits
The audit asks one question of every page: why would a language model treat this as low quality? Ivan Sp audits pages for the patterns that cause AI systems and quality algorithms to downgrade them. The audit asks one question of every page: why would a language model treat this content as low quality?
The Patterns the Audit Looks For
One audit of a rental listings article produced a 50-point itemized issues list. The findings fell into eight recurring categories:
- Promotional bias and advertorial tone in place of factual description
- Inconsistent or outdated date references mixed together in one article
- Vague statistics with no attribution and no source
- Thin content padded with generic lists to reach a length
- Leftover HTML and shortcode fragments visible in the rendered page
- Content that does not match what the title promises, such as a title offering 50 examples above an article delivering far fewer
- Off-topic tangents that dilute the page’s subject
- Individual filler sentences carrying no information
Findings get delivered as an itemized list rather than a summary, so each issue can be fixed and checked off separately.
Pass, Borderline, or Fail Scoring
Every page receives a status against ten criteria. Five are user-first indicators: evidence of first-hand experience, specific methods rather than general advice, likely reader satisfaction, actionable information, and original insight.
Five are red flags: promotional language, writing that exists to hit a word count, content that exists only to rank, generic advice available anywhere, and obvious keyword optimization.
The balance between the two sets produces a status of pass, borderline, or fail. Borderline pages usually justify a rewrite, and failing pages often justify removal instead.
Superlatives and Unsupported Claims
Unsupported superlatives get flagged individually. Phrases such as “top provider,” “the only ones who specialize,” and similar constructions state a claim the page cannot substantiate.
Bare numbers get the same treatment. A page claiming 87 industry awards without a source or a verifying link carries a credibility risk, and unqualified guarantee language with no stated warranty terms carries a legal one. Neither gets deleted automatically. Both get flagged for the client to substantiate or remove.
The Cost Content Gap
One finding appears often enough to name separately. Pages targeting commercial queries frequently omit pricing entirely, despite cost being among the first things a searcher wants.
On one kitchen renovation page, the absence of any price range was a larger opportunity than every on-page fix combined. Answer engines face the same problem, because a page with no cost information has nothing to offer a question about cost.
What Happens to Failing Content
Failing pages get sorted rather than deleted in bulk. Pages earning impressions get rewritten, because demand already exists. Pages earning nothing after a 60 to 90 day window get pruned or set to noindex.
Setting a page to noindex preserves it for later improvement while stopping it from diluting the site’s quality signal. One engagement marked 43 pages noindex in a single pass across two sites. The full method sits under AI search optimization as the parent service.
Audit One Page in Ten Minutes
I suggest a scan that finds the most common failures quickly.
- Open your most important commercial page and use your browser find function.
- Search for the words leading, fastest, and always.
- Search for the word studies, and for any percentage sign.
- For each hit, ask whether you could produce the evidence today.
Most pages return several hits and no evidence for any of them. Deleting those specific sentences takes minutes and removes every one of the exposures.
Where the Standard Content Audit Advice Breaks Down
Is there a claim on your site that you could not defend if somebody asked?
- The usual advice
- Score content against a quality checklist, then improve whatever scores lowest.
- What holds up in practice
- A checklist score aggregates unrelated faults into one number. An unsupported superlative and a missing cost section need completely different work, and averaging them hides which page is actually a liability.
Picture an audit reporting quality scores between 4 and 8 across sixty pages. The team starts at the bottom and works upward.
One page scoring 7 carries an unsourced statistic that a competitor could challenge publicly. The scoring treated it as a minor deduction against an otherwise strong page.
Ivan Sp grades pass, borderline, or fail and names the failing sentence. A liability is a fail regardless of how good the rest of the page is.
An averaged score buries the sentences that carry real risk. The page reads well, the number looks acceptable, and the exposure stays live.
One unsourced figure is all it takes for a sceptical buyer to stop believing the rest of the page. Nothing in your analytics will record that moment.
Four Things Commonly Assumed About Content Quality
Which of these appear on your own pages?
I meet these four assumptions on most content quality reviews.
| Commonly assumed | What actually holds |
|---|---|
| Quality is subjective. | Several failure patterns are objective. An unsourced statistic, a circular definition, and a contradiction between pages are all checkable. |
| Superlatives make copy persuasive. | An unsupported superlative invites a challenge the page cannot answer. A specific, checkable claim persuades a sceptical reader further. |
| A quality score identifies what to fix. | An aggregate score averages unrelated faults. Naming the failing sentence is what makes an audit actionable. |
| Longer pages score better. | Length is unrelated. A short page with sourced claims outperforms a long one built on vague quantifiers. |
| Marketing copy is exempt from this. | Marketing copy is where unsupported claims concentrate. A sceptical buyer applies the same test a machine does. |
Every unsupported claim on your site is sitting there waiting for the one reader who checks. Removing it costs nothing and it is the cheapest credibility available.
A Worked Example: Five Sentences Graded
Consider five sentences from one page, graded against the audit. The assessment below is illustrative, written to show what each grade means. No client engagement is described here.
| Sentence pattern | Grade | Why |
|---|---|---|
| The foremost roofing contractor in the region | Fail | Unsupported superlative, and no evidence exists |
| Studies show most roofs fail early | Fail | Unattributed statistic with no checkable source |
| We always deliver on time | Fail | Absolute claim admitting no exception |
| Roof repair is the process of repairing a roof | Fail | Circular definition, telling a reader nothing |
| Flat roof inspections take two to four hours | Pass | Specific, checkable, and bounded |
I name the sentence rather than scoring the page. Four failures on one page is a rewrite brief, not a score of six.
An unsupported superlative is a receipt with the total handwritten. Everything else may be accurate and nobody trusts the document.
The passing row is shorter than any of the failures. Specificity is usually fewer words rather than more.
How a Quality Audit Runs
I read for specific failure patterns rather than for a general impression.
- List every superlative on the page and check whether evidence exists for each.
- Remove every superlative that cannot be supported, rather than softening it.
- Flag every absolute claim and rewrite it with an honest qualifier.
- Find every statistic and check whether a source is named and reachable.
- Remove or source every unattributed figure before publication.
- Replace vague quantifiers with a number wherever a number is actually known.
- Check each definition for circularity, since a term defined using itself explains nothing.
- Compare facts stated on more than one page and resolve any contradiction.
- Confirm the page addresses what governs cost, since silence on that reads as evasion.
- Grade the page pass, borderline, or fail, and name every failing sentence.
How This Engagement Is Scoped
I scope quality audits by page count, since the work is sentence-level.
- What arrives
- A pass, borderline, or fail grade per page, with every failing sentence named and rewritten.
- Typical duration
- One to two weeks for a set of thirty pages.
- What the client supplies
- The page list, and the evidence behind any claim worth keeping.
- Included
- Pattern audit, grading, sentence-level findings, and rewritten replacements.
- Not included
- Legal review, and sourcing research for claims the business cannot evidence.
- What changes the estimate
- Large page sets, or heavily promotional copy needing substantial rewriting.
- How it is quoted
- After a short scoping call, once the page list is agreed.
Tools and Primary Sources
The references below define the quality standards behind the audit. Each opens in a new tab.
- Creating helpful, people-first content sets out the self-assessment questions this audit operationalises.
- Search Quality Rater Guidelines describes how raters assess substantiation, expertise, and trust.
- Google spam policies covers the patterns treated as manipulation rather than as weak writing.
- Google AI features documentation describes how AI surfaces select and attribute the content they draw on.
Common Questions About AI Content Quality Audits
Is this the same as checking whether content was AI-generated?
No. The audit measures whether content displays low-quality patterns, regardless of who or what wrote it. Human-written marketing copy fails these checks routinely.
How many pages can be audited at once?
Site-wide scoring runs through a crawl pipeline that handles 50 to 2,000 pages, scoring each one for helpfulness, originality, legal risk, and spam signals. Detailed itemized audits run on individual high-value pages.
Does removing promotional language weaken the marketing?
It changes what the marketing rests on. Specific factual statements outperform vague superlatives with both readers and quality systems, and a claim the business can substantiate is more persuasive than one it cannot.
Is removing superlatives the same as weakening the copy?
No. An unsupported superlative persuades nobody who is paying attention. A specific, checkable claim in its place is stronger with exactly the readers worth convincing.
Discuss This Work
The first call is a free 30-minute consultation with no pitch attached. You receive an honest assessment of where the site stands and what can realistically be improved.