Field note
AI Content at Scale: Avoid Search Spam
Publish only when each page has a distinct audience, evidence, owner and reason to exist.
Publish only when each page has a distinct audience, evidence, owner and reason to exist.
What the risk is
Google's spam policies include a category called scaled content abuse: producing many pages mainly to manipulate search rankings, without adding value for the people who land on them. The policy is explicit that it does not matter how the pages were made. Automation, AI, human writers working from a template or a mix of all three can all fall foul of it. Google's earlier guidance on AI-generated content says the same thing from the other direction: AI content is not against the rules in itself, and the question is whether the page is helpful, reliable and made for people.
For a small business the risk is practical, not theoretical. Pages that are flagged as spam can be removed from results or the whole site can be demoted. The pages that took a weekend to generate can take months to recover from, and the recovery starts with deleting them.
How small businesses get here
The pattern is nearly always the same. A tool makes it cheap to generate a page, so someone multiplies. A plumber produces "emergency plumber in [town]" for two hundred towns. An accountancy firm produces a "guide to [tax topic] for [industry]" grid. A software company produces a comparison page for every competitor it has heard of. Each page changes the town, the industry or the competitor name, and nothing else of substance.
Nobody at the firm thinks of this as spam. They think of it as coverage. But the reader who arrives on "emergency plumber in Kettering" and finds the same text as Corby, with no Kettering phone number, no Kettering engineer and no Kettering job, has learned nothing, and Google's systems are built to notice that.
A realistic example
A five-person electrical contractor in Kent used an AI tool to generate 180 location pages and 60 service pages in a month. Traffic rose for six weeks, then fell below where it started. Search Console showed most of the new pages indexed and then dropped, and the firm's original service pages fell with them.
The fix was to delete all but nine location pages, one for each area where the firm actually had a van and recent jobs, and to rewrite those nine with real evidence: the engineer covering the area, three recent jobs described in a sentence each, the response time they could actually promise. The site recovered over about four months. The lesson was not "do not use AI". It was "do not publish a page you could not stand behind in a phone call with the reader".
A test for every page before it goes live
Ask these five questions. A "no" to any of them is a reason not to publish.
- Distinct audience. Is there a real group of people who would search for this and not be served by an existing page on the site?
- Distinct evidence. Does the page contain something specific to its subject that the template did not supply: a photo, a job, a price, a named person, a local fact you checked?
- A named owner. Will someone in the business read it, answer for it, and update it when it goes stale?
- A reason to exist without search. Would you send this page to a customer in an email? If it exists only to be found, it exists only to be flagged.
- Honest scale. If you are publishing more than ten pages of the same shape at once, can you say why each one passed the four tests above?
Controls that keep AI useful
Use AI for the parts it is good at: turning notes from a site visit into a clear paragraph, drafting a first version for a person to correct, checking readability. Keep the facts, examples and local detail as human input, and treat the model's fluent filler as something to remove rather than something to publish.
Keep a publishing log with the page, the owner, the evidence used and the date reviewed. Set a limit on how many pages of any one template the site will carry, and review the ones that get no traffic after three months with a view to consolidating or removing them.
If you already have a large set of near-identical pages, the safer route is usually to consolidate them into fewer, better pages with redirects rather than to try to "differentiate" each one with more generated text.
What to write into the content policy
- Every published page has a named owner and a recorded evidence source.
- Template-generated pages require a per-page justification, not a batch approval.
- AI drafting is allowed; AI-only facts, quotes and statistics are not.
- Pages with no traffic and no evidence after a quarter are consolidated or removed.
- Search Console is checked monthly for manual actions and indexing drops.
Our notes on Google AI search visibility for UK SMEs and AI search content that earns citations cover what does work: fewer pages, each with something a machine could not have made up.