The content is just AI output nobody checked

There is a version of content marketing that costs a few hundred dollars a month and cannot possibly work, and it is easy to miss because nothing looks broken. The posts arrive on schedule. They are grammatical, sensibly structured and pleasant enough to skim. They are also interchangeable with the posts on your nearest competitor's site: no price, no suburb, no step in how you actually do the job, no story about the one that went wrong.

This is failure number five in my series on why SEO agencies fail, and the one clients defend longest, because content feels like the part they are definitely getting. Twelve posts on the blog are visible in a way a title tag fix is not. Visible and valuable are different things.

This is not an argument against AI

I use AI heavily in my own work. It drafts for me, it fixes things I have written badly, it saves me hours a week. I have a whole series on doing that properly, and I am not about to tell you the tool is the problem.

The failure is the missing layer. Generating a draft takes ninety seconds. Making it worth publishing takes a person who knows what you charge, which jobs are profitable, what customers ask on the phone, and which competitor is beating you. Without that person you get output rather than content, and the client pays for it.

There is usually an innocent explanation too. Whoever produces your posts has probably never spoken to you, may be carrying thirty accounts, and works from a one-line brief saying "plumbing, Christchurch, four posts a month". Nobody in that chain is dishonest. They were asked for a countable thing and they are producing it. That is a design problem, not a character problem.

Generic content fails twice, and the second failure is the expensive one

It does not rank, because it adds nothing to what is already indexed

Google already holds thousands of versions of "how to choose a plumber" or "five signs you need a physio", nearly all written by averaging the ones before. When your post joins that pile, the only honest question is whether there is a reason to show yours instead of one that has been there six years with links pointing at it. Competent is not a reason. Competent is the baseline everything on page one already clears.

This hits hardest the businesses least able to absorb it. A big brand can publish forgettable content and still rank, because the domain drags it up. A local service business has no such buffer. Its only real lever is being genuinely more useful on a narrow topic, and useful means information not available elsewhere. Strip the specifics out and that advantage goes with them.

It does not persuade, because it describes nobody

This is the failure that costs money, and almost nobody measures it. Say a post does pick up traffic. Somebody with a real problem lands on it, reads two paragraphs of "regular maintenance is key to avoiding costly repairs", and gets nothing: no sense of cost, no sign the business has done this before, no answer to what they were anxious about. They leave. No report ever says "forty-one people left because the page sounded like nobody wrote it".

I have written separately about why AI content sounds generic and the tells readers pick up without being able to name them. People are far better at detecting hollow writing than at explaining it. They quietly decide you are not the one, and ring somebody else.

Ranking and persuading need the same ingredient - information only your business could have supplied. That is why generic posts fail at both ends rather than one.

The ten-minute self-check: the highlighter test

No SEO knowledge needed. It is the fastest way to find out whether a human who knows your business was ever involved.

1. Open the last three posts published on your website. Not your favourites - the three most recent, whatever they are.
2. Get a highlighter, or use the highlight tool in a document. Read each post top to bottom.
3. Highlight every sentence containing a fact only your business could have supplied: a real price range, a specific job, a step in how you actually work, an objection a customer raised last week, a part you refuse to fit, a suburb or building type or council rule.
4. Be strict. "Our team has years of experience" is not such a fact - any competitor could write it verbatim. "The 1930s bungalows in Mount Eden nearly always have the same gib-lined ceiling problem, and it adds about two hours" is.
5. Count the highlights across all three posts, then do the same to a competitor's post and see whether the two piles differ.

If nothing lights up across three posts, that tells you something specific: nobody with knowledge of your business was involved. Not that the writer was lazy, not that AI was used - the raw material never got into the room. Content made that way is unlikely to earn a ranking and will not persuade the people who do read it. One highlighted sentence across three posts is a garnish, not a pass - and because the check needs no tools or logins, you can run it on a prospective agency's client sites before signing anything.

Why a volume target in the contract produces this structurally

This part is worth understanding properly, because it explains why swapping agencies rarely fixes it. "Four blog posts a month" is a countable unit: scoped, priced, scheduled, invoiced, ticked off, and everyone can verify it happened. "One post a month containing things only this client knows" is not countable that way. It needs a conversation with you, judgement about what is worth saying, and somebody senior enough to decide a draft is not good enough yet. None of that fits on a delivery board.

Whatever the contract counts is what delivery optimises for. That is not cynicism, it is how any production system behaves. If a good month means four posts, every minute spent chasing you for a story about a real job puts the count at risk. The writer is not choosing generic over specific. They are choosing on time over late, which is what they are assessed on.

The economics point the same way. Work out what share of your fee covers those four posts and divide by four. That number pays for research, writing, editing, images, uploading and internal linking. On most retainers I see it buys about half an hour of a junior person's time per post - enough to generate, skim and publish, nowhere near enough to interview you. You are getting the honest output of the time that was bought, and I have gone through those numbers in the hidden cost of AI-generated articles at scale. The fix is not a better writer. It is changing what the contract counts.

The cannibalisation risk of publishing near-identical posts

There is a second-order problem with volume that rarely comes up when the package is sold. Cannibalisation is the plain-English problem of having several pages on your own site all chasing the same search. Google will not rank them all. It picks one, and because they are so similar it is not confident which, so it keeps changing its mind and rankings wobble between three pages without ever settling.

Generic content makes this far likelier, because it converges. Take the specifics out and every post about emergency call-outs, urgent repairs and after-hours service is the same document with different headings. Twelve of those a year is not twelve assets. It is one topic split twelve ways, with authority and internal links spread thinly instead of concentrated on the page you want ranking.

It also makes the site harder to fix later. Brought in afterwards, the honest recommendation is usually to merge most of it into a handful of useful pages and redirect the rest, so much of two years' spend gets consolidated away. I have covered that in the cannibalisation trap in AI content buildouts. Put bluntly: publishing nothing would often have left the site in better shape.

What good content process looks like, and what you can supply

Done properly, step by step

The difference is not talent or effort but where the decisions get made. On a schedule: pick a topic off a list, generate, proofread, publish, invoice. Done properly it looks more like this.

• The topic is chosen because a specific search exists and no page owns it yet - not because the calendar said Tuesday.
• Before anything is written, somebody decides which existing page the post must not overlap with and which two pages it links to. That decision prevents cannibalisation rather than cleaning it up later.
• The brief contains raw material from the business: a price range, a real job, a customer objection, a photo, a constraint. Somebody had to go and get that, which is the step that gets skipped.
• AI does the drafting, and it should - that part is typing and it is fast. The material is what makes the draft worth anything.
• Someone who understands the business reads the draft against the highlighter test before it goes live and sends it back when it fails. Sending drafts back is normal, not a crisis.
• A few months later somebody checks whether the post earned anything, and that answer informs the next topic choice.

Notice how little of that is writing. Most of it is deciding and checking, which is exactly what a volume contract has no budget for. For the version aimed at doing this yourself rather than judging what you are handed, see my guide to using AI for your website content without wrecking it.

Twenty minutes that change the output

Clients underestimate how little raw material this takes. You do not need to write anything - just talk for twenty minutes and let somebody else do the rest.

• A rough price range for your three most common jobs, and what pushes one to the top of that range rather than the bottom.
• Three recent jobs, described the way you would describe them to a mate: what the customer thought was wrong, what actually was, what it took to sort.
• The three questions you get asked on nearly every call, and the answers you give.
• The objection you hear most often and what you say to it. The most valuable thing on this list, and almost nobody supplies it.
• Something you refuse to do, and why. Refusals persuade hard, because no competitor's generic post contains one.
• Anything local and concrete: housing stock in the suburbs you work in, a council requirement, a supply problem, a seasonal pattern.

A voice memo recorded between jobs covers most of it, and whoever writes for you can work off that transcript for months. If you supply all that and the content still comes back interchangeable, you have learned something: the material was not the blockage.

How to renegotiate the content line item to fewer and better

You do not need a confrontation or a cancellation, just a change of unit: same budget, one post a month instead of four, the difference spent on a twenty-minute call with you and an editing pass by somebody senior. Worth asking for:

• One post a month, with a written brief you see first. If you cannot tell from the brief that it is about your business, the post will have the same problem.
• A standing twenty-minute call, or a shared document where you drop jobs and questions as they arise.
• The highlighter test as an explicit acceptance standard: you will not sign off a post containing nothing only your business could have supplied.
• Which search phrase the post is meant to win, and which existing page it must not compete with. Any provider with a content plan answers that in a sentence.
• A six-month check on what the posts actually earned, with the option to stop the line item if the answer is nothing.

A provider working properly finds this easy, and some will be relieved, because producing content nobody reads is no fun at their end either. One who resists, or who explains that volume is what search engines reward, has told you which model they run. It is one of the sharper things to probe when choosing an SEO provider. If they will not move, run it from your side anyway: same fee, fewer posts, raw material from you. A lower count is not a loss.

The short version

Content does not fail because AI wrote it. It fails because nobody made the judgement calls - what to write about, what to say that nobody else can say, and whether the draft was good enough to publish. Those calls need a human who knows your business, and no volume target has ever produced one.

Go and run the highlighter test on your last three posts. Ten minutes, and no ambiguity in the result. If nothing lights up, the problem is not the writer's talent and not the tool. Your business was never in the room - which is a fixable thing to raise on your next call.