Some were drafted by your team, some by freelancers, and some arrived from a client with no notes on how they were made. Most are fine. A few are not, and you would rather find them yourself than hear about them from a client or watch them slide down the rankings.
This is a working playbook for that job. It covers what Google actually says about AI content, how to check your website for AI content with a five-phase audit you can run on any site, a tracker template, and a short script for explaining the results to a client.
Start with what Google actually says
A lot of audit anxiety comes from a misreading of Google's AI content guidelines, so it is worth getting them right before you touch a single page.
Google does not ban AI content. In its official guidance on AI-generated content, Google Search says it rewards high-quality, helpful content however it was produced, and that using automation mainly to manipulate rankings is what breaks the rules. The focus is on usefulness, originality, and signs of real experience and expertise.
The line that matters most for marketers sits in Google's spam policies, under scaled content abuse. Google defines it as producing many pages mainly to manipulate rankings rather than help users, and it lists using generative AI tools to create many pages without adding value as one example. Scraping, spinning text with synonyms, and pages stuffed with keywords sit in the same category.
So the goal of an audit is not to remove every trace of AI. It is to find pages that read generic, add nothing a reader could not get elsewhere, or look like part of a mass-produced batch. A style score is useful here as a proxy: pages that read strongly machine-made are often the same pages that lack original detail. Google has not said it uses AI detectors, so treat the score as a way to find weak content, not as a prediction of rankings.
Why audit at all?
-
Clients ask. More contracts now include clauses on AI use, and some clients run their own checks on delivered work.
-
Generic pages underperform. Text that says what every competitor says rarely earns links, shares, or conversions, whatever its rankings.
-
Scale creates risk. One thin AI page is a quality problem. Two hundred of them, published in a month, start to look like the pattern Google's spam policies describe.
-
Freelance work varies. An audit gives you a consistent content quality check across writers, and an AI content checker for SEO teams makes that quick to repeat.
The five-phase AI content audit
Phase 1: Inventory
Goal: a list of pages worth checking, ranked by how much they matter.
Export your indexed URLs from Google Search Console or your CMS, then add clicks or sessions from the last 90 days and a simple business-value rating. A product page that drives leads matters more than a three-year-old news post with no traffic. For large sites, start with the top 50 to 100 pages by traffic plus anything published in the last six months.
Phase 2: Scan
Goal: an AI-likeness score for each priority page.
This is where you check your website for AI content, page by page. You do not need to copy and paste every page. Some tools let you scan a URL for AI text directly: you paste a public page address, the tool pulls the readable text, and it returns a percentage with the most machine-sounding lines highlighted. That works for homepages, articles, and service pages. It will not get past login walls, and PDFs usually need their text pasted in by hand. Long pages may need to be checked in sections.
Whichever AI content checker for SEO work you use, stick with one tool for the whole audit. Different detectors give different numbers, so mixing them makes your results impossible to compare.
Phase 3: Triage
Goal: a clear action for every page.
Plot each page on two axes: how much it matters to the business, and how strongly it reads as AI-generated. That gives you four groups.
- Fix first: high value, reads AI-like. These pages carry the most risk and the most upside.
- Protect and watch: high value, reads human. Keep them current and re-scan after major edits.
- Rewrite or merge: low value, reads AI-like. Often thin posts that cover the same ground as other pages.
- Leave as is: low value, reads human. Not worth your time this round.
Before you act on a score, read the page yourself. A high score on a technical spec page or a legal disclaimer may simply reflect formal, repetitive language, and that is not a problem.
Phase 4: Fix
Goal: pages that offer something a reader could not get anywhere else.
The fix depends on the group, but on most "fix first" pages it means adding what only your business can provide:
- Real numbers, case examples, screenshots or customer questions
- A named author with relevant experience, and a short line on how the page was put together
- A clear point of view instead of a balanced summary of what everyone else says
- Updated facts, prices, and dates
For stiff passages, an AI rewriting tool can loosen the rhythm as a first pass, but keep the keywords you need and do the final edit by hand. A rewrite changes how text reads. It does not add the expertise that makes a page worth ranking.
For the "rewrite or merge" group, combining three thin posts into one strong page and redirecting the old URLs usually beats polishing each one. Pages with no value and no traffic can be removed or set to noindex.
Phase 5: Monitor
Goal: stop the problem coming back.
Add a pre-publish step to your workflow: every new page gets a quick check before it goes live, the same way it gets a proofread. Then re-run the audit on your top pages every quarter. If you use AI writing for blogs at volume, also track how many pages you publish each month, since a sudden spike is the pattern most likely to draw attention.
What to tell the client
Clients rarely want a lecture on detection methods. They want to know whether there is a problem and what you are doing about it. A short update works best:
"We checked your top 60 pages for content that reads generic or machine-written. Twelve need work, and five of those are high-traffic pages, so we are starting there. The fixes add your own data, examples, and expert input. Google doesn't penalise AI use on its own, but it does favour pages that are clearly useful and original, so this should improve quality whichever way the pages were first drafted. We'll re-check new pages before they go live."
Adjust the numbers, keep the tone calm, and avoid promising ranking gains you cannot guarantee.
Where audits go wrong
-
Chasing a 0% score. Formal, well-structured human writing often scores above zero. Aim for useful pages, not pages that pass a number.
-
Treating the score as proof. If a freelancer's work scores high, review it with them before concluding.
-
Rewriting everything at once. Bulk changes across hundreds of pages make it hard to see what helped. Work in batches and watch the results.
-
Ignoring the pages that score low. A page can read human and still be thin, outdated, or wrong.
Make it a habit, not a project
The sites that never need a big clean-up are the ones that check as they go. Build a check into every publish, re-scan the pages that matter each quarter, and keep asking the question Google's guidance keeps coming back to: would a reader be glad they found this page? If the answer is yes, how the first draft was written matters far less.
