Automation
4,000 product pages to write: what genuinely gets automated, and the trap that tanks Google
An online retailer calls me one evening, with the voice of someone who has had enough of his own catalogue. He sells spare parts online, several hundred references, and more than half his product pages are either empty or copied word for word from his supplier's site. His summary fits in one sentence: "I sell, but on Google, I do not exist." He was not wrong, and the problem was not where he thought it was.
A page copied from the supplier, for Google, does not exist. The engine sees the same text on thirty sites and rightly decides it has no reason to put yours ahead of the others. An empty page is worse: it announces there is nothing to read. His catalogue was full, his shop was running, but in Google's eyes it was half transparent. Here is what we automated, and above all the exact spot where the machine could not be left to work alone.
Why copying the supplier gets you nowhere
When I open a shop in this state, I always start the same way: I take a sentence from a product page, paste it in quotes into Google, and look. Nine times out of ten, the same description turns up on fifteen other sites, word for word. It is the fastest demonstration I know. If your text already exists elsewhere identically, you are not fighting for first place, you are out of the race before the start. He watched it happen live on his own screen, and it settled the "does content actually matter" debate in three minutes.
What we put in place, and what we kept under control
The idea was never to "generate text". It was to start from what he already knew about his parts (what they are for, which models they fit, what makes a customer choose this one over another) and let the machine put it all into shape, page after page, in a template built together with him. He checks, adjusts, publishes. What used to take him twenty minutes per page on the rare occasions he forced himself to do it, work he kept putting off as a result, became a quick read-through of a batch prepared overnight.
Eight hundred pages brought up to standard, checked in a single afternoon of review, instead of months of writing kept being put off evening after evening.
The stumble: at first, every page sounded the same
The first batch was clean, correct, and perfectly interchangeable. Every page followed the same mould, the same turns of phrase, the same way of saying "ideal for". Read one after another, they reeked of machine, exactly what we wanted to avoid. We went back over the whole template to inject his own way of talking about his parts back into it: his words, his comparisons, the advice he slips in on the phone to a hesitant customer. The pages started sounding like someone again, not like a form. This is not an optional step. It is the one that separates "content" from "your content".
The real trap: automating on top of wrong data
Where it turns dangerous is when you put page production on autopilot without having checked what lies underneath. On another shop, I once found pages that were well filled in, well written, selling products of which not a single unit was left in stock. Had we plugged automatic generation on top of that flaw, we would have produced beautiful pages for items impossible to deliver, at scale, and a stream of customer service complaints to match. It is the principle I repeat everywhere: automating a process that is already broken reproduces the same mistakes, only faster and at greater scale. I tell more stories like it in the three levels of AI in a business, because it is the trap that costs the most.
What Google actually wants
We make search ranking sound complicated, but the starting idea is simple. Google wants to send its users to the page that best answers their search. A page that genuinely describes the product, with the details a buyer is looking for (dimensions, compatibility, what it is for, what sets it apart from the one next to it), answers better than an empty page or text copied elsewhere. You are not writing "for Google", you are writing for the buyer, and Google rewards that because it is exactly in its own interest too.
That is why automation, here, is not a machine that fills emptiness with words. It is a machine that puts into shape, quickly and at scale, real product knowledge that already existed in the seller's head but showed up nowhere on the site. The value comes from him; the machine only makes it visible, page after page. The day you flip that around, asking the machine to invent knowledge you do not have, you fall straight back into the hollow text that no one reads and Google does not rank.
What you can do alone this week
You do not need anyone to run this check. Take your ten best-selling products and run the quote-marks test in Google, one page at a time. Note the ones whose text appears elsewhere identically, and the ones that are flat-out empty. That list is your priority job: it is your best sellers working at half capacity.
Producing at scale, with a template that keeps your voice, is where we step in. If you are redoing your shop at the same time, read my approach to a redesign that does not vanish from Google first, because the two projects often overlap. And for the detail of what we set up on writing and search, it is all on my dedicated content page.