Automating website content without it reading like spam

What Google actually says

There’s a lot of folklore about Google and AI content. Google’s own guidance is shorter and clearer than most of it.

Its guidance on generative AI content says AI “can be particularly useful when researching a topic, and to add structure to original content”. So using AI isn’t the problem.

The same page warns that using AI or similar tools to generate many pages without adding value for users “may violate Google’s spam policy on scaled content abuse”.

And it’s blunt about accuracy: it’s “critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing”. That applies to titles, meta descriptions, structured data and image alt text too, not just the visible text.

What scaled content abuse looks like

Google introduced the scaled content abuse policy in its March 2024 spam policy update. Broadly, it’s producing lots of pages mainly to rank, with little value to the people who land on them. Typical examples:

• pages generated in bulk with little or nothing to tell them apart
• articles rewritten from other sites, by hand or by machine
• keyword-stuffed pages that don’t really answer anything

None of that depends on whether a person or a machine wrote it. What matters is whether the page is useful.

The approach that works: start from your own data

The pages worth automating are ones built from information you already have and nobody else does. For example:

• Products: specifications, sizes, compatibility, stock, prices.
• Locations: branches, service areas, opening hours, what’s available where.
• Your own records: projects completed, questions customers actually ask, prices you actually charge.

A page built from real data has something a rewritten article can’t: facts specific to your business. AI is useful for turning that data into readable text, structuring it consistently and spotting gaps. It shouldn’t be inventing the facts.

What my own directory sites taught me

I run two directory sites built this way, so I can show you real numbers rather than theory.

HomeKit Directory: scale isn’t the same as value

HomeKit Directory lists smart home products that work with Apple Home. It has over 1,200 product pages, generated from product data.

A review of 90 days of Search Console data found that 580 of those pages had not appeared in Google once. Not a single impression. They’ve since been taken out of Google’s index, because a site where half the pages earn nothing doesn’t look like the best resource on anything.

What does rank is telling. Not the head terms: “homekit thermostat” sits around position 70. The pages that reach Google’s top ten are obscure products that nobody else documents properly: a particular ceiling light, a garage door controller, a couple of wall switches. Coverage that doesn’t exist elsewhere is the site’s real edge.

The other lesson was the copy. Many product descriptions came in from the import, written in the “Experience the future of home automation… Order yours today!” style. That’s exactly the kind of template text Google’s policy describes, and it was feeding the meta descriptions too. The fix is to rewrite it from verified product data, which is next on the list.

Manchester Photo Spots: structured facts, written for people

Manchester Photo Spots is a guide to photography locations in Manchester. Each of its 77 location pages is built on real details: the best light, the nearest tram stop, parking, access, and the coordinates for the map. Parts of each page, like the shooting tips, are generated from those details, and can be overridden by hand where there’s something better to say.

In the three months to 1 September it earned 37,200 impressions in Google, at an average position of 14.9, up from 25.9 earlier in its history.

It also showed the limits. A page answering a factual question, such as how many people a venue holds, can rank on page one and still get very few clicks, probably because the searcher gets the answer without needing to click. Ranking isn’t the same as visitors.

What both have in common

Both sites have almost no links from other websites, and that’s now the ceiling on both. Good structured pages get you into the results. Getting near the top for competitive searches takes other sites pointing at yours, and no amount of automation does that for you.

The checks before anything goes live

If you’re generating pages in bulk, build the checks in:

1. Every fact comes from your data. If the model writes a number or a claim that isn’t in the source, it’s wrong, however plausible it sounds.
2. Each page answers a real question. If two pages would say the same thing, publish one.
3. A person reviews a sample. Read a proper selection of pages before publishing, and again after any change to how they’re made.
4. Check the metadata too. Titles and descriptions appear in search results. They need the same care.
5. Say how it was made, where it helps. Google suggests considering explaining how content was created in a way that makes sense for your audience.

What not to automate

Some content needs a person: opinions, advice, anything where your experience is the point. A model can draft and tidy, but if the article is meant to show what you know, you have to be the one who knows it.

Where to start

Look at the information your business already holds in spreadsheets, systems and documents, and ask: is there a question customers search for that this data answers? If so, that’s a page worth building, and probably many more like it.

If you’d like help working that out, it’s one of the things an audit covers.