Publishing content automatically is now easy. Knowing whether any of it is working, or quietly damaging the site, is the part nobody has a process for.
Google's position is that automation is fine and low-value content is not, regardless of how it was produced. So the audit is not "was this written by AI". It is whether each page is accurate, distinct, matched to intent and useful, which is the same standard as for anything else. We answer the prior question, whether automated content actually works for blog SEO, separately and without the sales pitch.
This is the nine-step process, ordered so the checks that find real damage come first.
Before you start
Pull three things: your full list of automatically published URLs, twelve months of Search Console data, and the publication date of each page.
Set a sample size. If you have published 500 pages, auditing every one by hand is not happening. Audit all pages with impressions, plus a random 10 percent of the rest, and treat what you find as representative. If sampling by hand is the bottleneck, an automated site audit crawls the full set and flags the pages worth reading.
The order, and why
| Step | Finds | Time per 100 pages |
|---|---|---|
| 1. Factual errors | Damage to credibility | 3 hours |
| 2. Fabricated entities | The worst failure mode | 1 hour |
| 3. Cannibalisation | Self-inflicted ranking loss | 1 hour |
| 4. Intent and format mismatch | Pages that cannot rank | 2 hours |
| 5. Thin and duplicate | Sitewide quality drag | 1 hour |
| 6. Indexation | Whether Google kept them | 30 min |
| 7. Performance triage | Where to spend the fixes | 1 hour |
| 8. Internal links | Orphans | 30 min |
| 9. Decide: fix, merge, delete | The output | 2 hours |
Step 1. Check the facts, starting with anything numeric
Every price, statistic, date, percentage and product capability. This is the most expensive check and the one that matters most, because a wrong number sits on your site for years looking exactly like a right one.
Prioritise: numbers in headings and tables, anything attributed to a named source, anything about a competitor, and anything with regulatory or financial consequences.
Do not use AI to check AI. A model asked to verify its own output will confirm its own invention. Verify against the primary source or not at all.
Step 2. Verify every named product, company and person
The characteristic failure of automated content is the confident mention of something that does not exist. A tool that was never launched, a company that closed, a feature that was discontinued. Preventing it starts upstream, and our guide to building an automated content strategy for startups covers scoping the topic list before anything gets generated.
How to check it at scale: extract every capitalised entity from the pages, deduplicate the list, and load each domain. A tool that 404s, times out or resolves to a parked lander does not belong in the article.
This check has real teeth. On our own site it caught four products being recommended across several pages that either no longer existed or had never existed. Every one of those pages was recommending them to readers.
Step 3. Find cannibalisation
Automated publishing at volume produces near-duplicate targeting, because the topic list was generated rather than checked against the site. Checking the topic list against what you already cover is the preventative version of this, which is the job of automated content planning.
In Search Console: filter by query and look at the pages column. Any query where two or more of your URLs appear is a conflict.
The faster version: export your page and query data, group by query, and flag every query with more than one URL. Sort by impressions and start at the top.
Step 4. Check intent and format against the live SERP
For each page with meaningful impressions, search its main keyword and look at what ranks. Is it listicles, guides, tool pages, product pages?
If your page is the wrong format, that is the finding. A 3,000-word guide on a query where the entire first page is listicles will not rank no matter how good the writing is. Rewriting the format is a bigger win than any on-page fix.
Count the entries too. If the ranking listicles run twelve to sixteen items and your automated page has five, it is not competitive.
Step 5. Find thin and near-duplicate pages
Thin: under about 800 words on a topic where competitors run 2,000, or a page that answers the query in one line and pads the rest.
Near-duplicate: run a similarity comparison across your automated pages. Anything above roughly 70 percent overlap with another page on your own site is a merge candidate, not a fix candidate.
Template repetition counts. If every page opens with the same three sentences and only the keyword changes, that pattern is visible across the whole set and drags all of it. Requiring a shared content brief with a specific angle per page stops that pattern at source.
Step 6. Check what Google actually indexed
In Search Console's Pages report, compare your published count to the indexed count.
"Crawled, currently not indexed" is the diagnosis. Google reached the page, evaluated it and chose not to index it. At volume, this is the clearest possible signal about the quality of the batch, and it is free.
A low indexation rate on automated pages is not a technical problem. It is a quality verdict, and no amount of sitemap resubmission changes it.
Step 7. Triage by performance, not by date
Sort every page by impressions over the last three months and split into four groups.
- Impressions and clicks: working. Update the facts and leave it alone.
- Impressions, no clicks: title and meta mismatch, or wrong format. Cheapest fix on the list.
- Indexed, no impressions after six months: not competitive. Merge or delete.
- Not indexed: delete or rewrite completely. Do not tinker.
Spend the effort on group two. Those pages are already earning attention and are one round of work from earning traffic. The same prioritisation drives a full website content audit across pages that were written by hand.
Step 8. Check for orphans
Automated publishing usually skips internal linking, which leaves pages reachable only from the sitemap.
Crawl the site and find pages with zero internal inbound links. Then add links from your relevant existing pages, with descriptive anchor text. Where the pages form a cluster around one topic, our guide to content hub SEO covers how those links should be arranged.
This is a common and cheap win. We have seen a batch of automated pages sit flat for months purely because nothing on the site linked to them.
Step 9. Decide fix, merge or delete, then actually do it
Every page audited gets one of three outcomes, recorded.
Fix when the page has impressions or a genuine unique angle. Correct the facts, match the format, add the internal links, rewrite the opening.
Merge when two or more pages target the same intent. Combine the best of each into one, then 301 the others to it. This recovers the split authority immediately.
Delete when the page has no impressions after six months, no unique angle and no merge target. Return a 410 or 404, or redirect only if there is a genuinely relevant destination. Redirecting everything to the homepage is not a strategy.
Record the decision and the date. Without a record, the next audit repeats the same work.
What to check by hand, always
Some things do not survive automation and no tool catches them.
The opening paragraph. Generated openings are the most predictable part of any automated page and the first thing a reader judges.
Anything a customer could act on. Prices, deadlines, legal thresholds, safety information, dosages, tax figures. If being wrong has consequences, a human reads it.
Claims about your own product. These get out of date fastest and are the most damaging when wrong, because they are the ones a buyer relies on.
Anything about a competitor. Both for accuracy and because a mistake here is the one someone will publicly correct you on.
How often to run this
Quarterly while you are publishing automatically at volume. Twice a year once the process is stable and the failure rate is low.
Immediately after any change to your generation setup, on the first twenty pages produced under the new configuration. That is the cheapest possible moment to catch a systemic problem.
Preventing it rather than auditing it
Most of what this audit finds is preventable at generation time. Verify entities before publishing rather than after, check the SERP format before writing, check the topic against your existing pages before adding it to the calendar, and insert internal links as part of publishing. Which steps are safe to automate and which are not is the subject of our guide to automated content marketing.
Distribb does those checks in the pipeline: topics are selected against your existing coverage and your Search Console data so the same intent is not targeted twice, and internal links are added as each article publishes rather than in a later cleanup. The generation side of that pipeline is our AI content writer, which is where those checks run before anything publishes.
Honest limitation: no automated pipeline, ours included, removes step 1. Facts still need a human, and anything with financial, medical or legal consequences needs a qualified one. Any vendor claiming otherwise is describing a product that does not exist.
If you want an outside perspective on this, content performance metrics covers it well.
What to do first
Run step 6 today. It takes ten minutes in Search Console, it costs nothing, and the indexation rate on your automated pages tells you immediately whether you have a small problem or a large one.
Then run step 2 on your twenty highest-impression pages, because a fabricated recommendation is the one failure that damages readers rather than rankings.
Our guides to the SEO audit checklist, automated SEO reports and content gap analysis for SEO cover the wider audit, the reporting and the planning side of this.