One of our own published posts had this sitting in the middle of it: “This completes the second section of your article. Please let me know if you would like any further adjustments.” Not a draft. Live, under our name, on a page anyone could land on. That sentence kicked off a full audit of this blog. If you are working out how to clean up AI-generated blog content on a site you own (or inherited), the honest starting point is that somebody has to open every post.
So we did. All 178. Here is what happened, including the parts that make us look bad.
How does a blog end up with 178 posts nobody reads?
Slowly, and with good intentions. This blog came from HubSpot as roughly 177 posts, built across several years and more than one owner. Twenty-one still carried the previous owner’s brand in the SEO title.
Then the 2023 cluster. Posts about BBQ joints, sushi bars, pizzerias, jerky retailers, jewelry shops, chiropractors. We build WordPress sites for small businesses. We do not write about jerky. Someone decided volume was a strategy, aimed a generator at a keyword list, and let it run.
Quick sidebar, in case this is new vocabulary. “AI-generated content” means text a language model wrote from a short instruction, in a minute, for roughly nothing. That works. What it hands you is a first draft with no editor attached and no knowledge of your business. That distinction is the whole article.
Nobody decides to publish garbage. They decide to publish frequently, and the only way to hit that cadence is to stop reading what goes out.
Why score every post instead of sampling a few?
Because the bad ones do not announce themselves in the title. Almost nothing looked wrong from the WordPress list view. Reasonable headline, real featured image, sensible category. The rot was three scrolls down.
So we opened all 178 and scored them against the questions Google publishes about helpful, people-first content. Would someone leave satisfied? Does this exist because a human had something to say, or because a keyword had volume? A row per post, one verdict each.
Here is the number the “audit your blog in five easy steps” posts leave out. At five minutes each, 178 posts is fifteen hours before you change one thing on the live site. Then the redirect map. Then the rewrites. If you also answer the phone and quote the jobs, those hours are not sitting in your week waiting to be used.
What are the four buckets for an AI content audit?
Keep, refresh, redirect, or delete. No maybes, no “revisit later” pile. Here is how our 178 landed.
| Verdict | Means | Count |
|---|---|---|
| Keep | Accurate, on-topic, written by a person with a point. | 38 |
| Refresh | Good premise, bad execution or stale facts. Worth a human rewrite. | 45 |
| Redirect | Not worth keeping, but the URL has equity or a near-duplicate exists. | 40 |
| Delete | No premise, no audience, no fixing it. | 55 |
What shipped was 52 posts trashed and 44 redirects written. Not the verdict numbers. The traffic check explains four moves, and the rest is the gap between a spreadsheet and a live site. I would rather show both than reconcile them into one story.
Which AI posts are unsalvageable?
When the problem is the premise, not the prose. You can fix bad writing. You cannot fix an article nobody should have assigned. The tells:
- The topic is off-niche. Nobody who needs a website finds us through jerky subscriptions, and no rewrite changes that.
- The examples are invented. If the “case study” company does not exist, the whole spine is fiction.
- It exists to hold a keyword. Ask what the reader now knows. If nothing, delete.
- It duplicates a better post. Two thin articles on one topic compete and both lose. Redirect one.
What did we find inside live posts?
The part I would rather not publish, which is why it is here. Live pages, real artifacts:
- The “please let me know if you would like any further adjustments” line from the top of this article. Public.
- Fabricated case studies comparing “BBQ Joint A” and “BBQ Joint B.” Neither exists.
- A section about gender neutral fashion pasted into the middle of an article about BBQ restaurant analytics. No transition, just there.
- Several posts that stopped mid-sentence.
- Orphaned citation markers pointing at sources that were never included.
- Three posts that openly stated AI authorship. One said the whole thing was written by an AI language model. Another was a guide explaining that we generate our content with ChatGPT and NeuronWriter.
Read that last one again. A published guide telling readers we did not write our own words. Deleted, obviously. It survived a platform migration with nobody reading it.
And look. If you have run posts through a generator and something about the result felt off in a way you could not name, that list is the name. It is not your prompt. A model writes fluent, confident, average text with no memory of your company, no stake in your reputation, and no way to flag what it invented. That gap does not close with a better instruction. It closes with a person who knows the subject, reads every line before it goes public, and answers for it when it is wrong. Fluency got cheap. Judgment is what you are buying.
Should you check Search Console before deleting anything?
Yes, every time. Here it changed the list. Before trashing anything, we pulled 12 months of Search Console data against the delete pile.
The archive going in: 22 clicks and 31.1K impressions over twelve months, average position 36.1. Past the third page across everything it surfaced for. Technically at the party, standing in the driveway.
It earned its keep. Three posts marked for deletion had real impressions behind them (570, 340, and 97), so they became redirects instead of dead ends. One had a human click, so it stayed live. Four saved by looking first. Deleting a page that quietly ranks is self-inflicted.
Redirect or let it 404?
Redirect when a genuinely relevant destination exists. A better article on the subject, or a service page answering the same question, earns a 301. Sushi bar marketing on a site that sells WordPress builds does not, and aiming it at the homepage just moves the dead end somewhere more annoying. A 404 is a clean signal. Google copes.
The problems an audit catches that have nothing to do with rankings
This part surprised me most. A password manager post still recommending LastPass as a default pick, years after the breach that changed a lot of minds. An analytics guide touring the Universal Analytics interface, shut off in July 2023. A tutorial with a PHP snippet echoing a $_GET value straight to the page, no escaping. Reflected XSS, taught as an example, on a web developer’s blog.
The ranking cost of that post was nothing worth measuring. The cost to a reader who pasted that snippet into a live client site is a different number, and not ours to pay. Stale content is a liability with a slow fuse, which is why re-reading what is published belongs inside ongoing site maintenance, next to backups, updates, and security patching.
Which brings up the question owners ask when money is tight: can we pause maintenance? No. It is the first line people cut because nothing visibly breaks the week you cancel. It breaks later, the month you learn the backup you assumed existed does not. Cut somewhere else.
What we are not claiming
We do not know if this worked yet. The cleanup finished recently, Search Console moves in months rather than days, and anyone showing you a recovery chart two weeks after a prune is showing you noise. No traffic lift here, no percentage, no “and then our impressions tripled.”
What we will measure: impressions and position across the survivors, clicks on the refreshed posts, whether the deleted URLs leave the index cleanly. When there is enough to say something true, this post gets the numbers, including if the answer is nothing changed.
What does the cleanup actually take, step by step?
Order of operations matters more than it sounds. The exact sequence we ran:
- Export your post inventory. URL, title, publish date, category. A spreadsheet is fine.
- Pull 12 months of Search Console data and join it to that inventory before you form opinions.
- Open every post and assign one verdict. Read them. No sampling, no skimming the first paragraph.
- Cross-check the delete pile against traffic. Anything with real impressions moves to redirect or refresh.
- Write the redirect map first. Old URL to new URL, every row filled in, before you trash anything.
- Delete, redirect, then resubmit your sitemap. In that order.
- Fix the intake problem. If nothing changes about how posts get published, you will be doing this again.
Step seven is the one everybody skips. The audit is the cleanup. The standard you publish against is the fix.
That list is complete and it is yours. What it is not is small. Every step fails quietly in its own way: a typo in the redirect map, a sitemap resubmitted too early, a delete pile nobody traffic-checked, forty rewrites you never find an evening for. It is work we take on while building and looking after a site for small businesses, and it goes faster with us because we made these mistakes on our own domain first. The pricing calculator gets you a ballpark in a minute.
FAQ
Does deleting AI content actually help SEO?
It removes a known negative, which is not the same as adding a positive. Thin, off-topic pages give search engines a worse picture of what your site is about, so clearing them out helps. It does not create demand for the pages you kept. Cleanup plus genuinely useful new content moves the needle. The second half is the harder half.
Should I delete posts or just noindex them?
Delete with redirects if you are confident the content is bad and not coming back. Noindex is right when a page has a reason to exist for humans but no business in search results. Using it to avoid deciding just hides the mess.
Can I fix AI content with a better AI prompt?
No, and I would rather save you the weekend of trying. A prompt changes the sentences. It does not change whether the article had a reason to exist, whether the company in paragraph four is real, or whether the advice still holds. Those calls take real knowledge of your business. We use AI daily. It drafts. It does not decide, and it never publishes.
What if the bad posts were written before I owned the site?
Does not matter. Your domain, your byline now. We inherited most of this archive and it was still ours to answer for. The upside: no ego attached, and handing it off costs nothing but an email.
How many posts should I expect to delete?
There is no correct ratio. Be suspicious of anyone who hands you one. We deleted roughly a third, because roughly a third was written for an audience that was never coming.
The short version
Publishing your own mess is uncomfortable, and it is the only version of this article worth reading. Plenty of people sell content audits. Few show you the sentence their own blog shipped with the quiet part left in.
So, concrete next step. Send the domain and a rough count of what is back there, guess if you have to. Tell us what you inherited, we will read it, and you get a straight answer on what is salvageable, what is landfill, and what it costs. Want a ballpark tonight instead? The pricing calculator is right there. Leaving it up is the only option that keeps charging you.