Content at scale without losing quality

The temptation

Someone on the team runs an experiment. A spreadsheet of 400 towns, a template and an AI model. By lunch there are 400 pages, one for each town, each saying almost the same thing.

It feels like leverage. It is usually the opposite. Google has a name for this pattern, and it isn't a compliment.

The question isn't whether to use AI for content. It's how to use it so that each page earns its place.

What Google actually says

Google's position has been consistent for a while. It rewards helpful, reliable content however it's made, and acts against content made mainly to manipulate rankings, however it's made.

Its spam policies describe scaled content abuse: producing many pages mainly to manipulate search rankings rather than help people. The examples it gives include using generative AI tools to produce many pages without adding value, scraping or rewording other sites' content, and stitching together material from other pages without adding anything.

The same policies describe doorway abuse, which includes pages built for individual towns or regions that simply funnel visitors to one destination. The 400-town experiment fits that description almost exactly.

Google's guidance on generative AI content is just as clear on the other side. It says AI can be useful for researching a topic and adding structure to original content. It also says it is critical to fact-check and review AI-generated content for accuracy before publishing, and that the same review applies to titles, descriptions, structured data and image alt text.

Its guidance on helpful content adds a useful test: why was this page made? If the honest answer is "to attract search visits," rather than "to help a reader," that's a warning sign.

Where AI genuinely helps

Used well, AI takes real time off a content project. The useful jobs are mostly the unglamorous ones.

• Research gathering. Pulling together what a vendor's documentation says, so the writer starts with the material in one place.
• Structure. Proposing an outline that covers the questions a reader will have.
• First drafts. A rough version an editor can reshape, rather than a blank page.
• Consistency checks. Flagging where one page contradicts another, or where a term is used two ways.
• Turning notes into prose. An engineer's bullet points become a readable paragraph, which the engineer then checks.

The pattern in all of these is the same. AI handles the gathering and the first shape. A person makes the decisions: what the page says, what it leaves out, and whether it is true.

That split is also what keeps the voice right. Pages drafted and published without an editor start to sound alike, across your site and across everyone else's. Readers notice, even when they can't say why.

What AI should not be trusted with is the facts. Models predict likely words, and a likely-sounding claim can be wrong. That's where the process has to slow down on purpose.

Fact-checking against primary sources

Every factual claim on a page should trace back to the source that would know. Not a blog that summarized it. Not another AI answer. The source itself.

• What a software product can do: the vendor's own documentation or help center.
• What a regulation requires: the regulator or the published rule.
• What a search engine does: its own documentation for site owners.
• What your company does: the people who do the work.

Link to the source on the page. Readers can check it, and so can the systems that decide whether to cite you.

Be strict about numbers. If a statistic can't be traced to its original study, cut it. A page with no statistics is better than a page with an invented one. Invented figures spread, and they spread with your name attached.

Record the date each page was checked, and put it on the page. Then actually recheck it. Vendors rename features, change plans and retire products. A page that was right last year can quietly become wrong.

The human editor

An editor isn't there to fix commas. They decide whether the page should exist at all.

A good editor asks what this page tells a reader that the other pages don't. They cut the paragraphs that say nothing. They replace general claims with specific ones. They make sure the page sounds like your company, not like every other page on the topic.

Then, for anything describing how your company works, someone who does that work reads it. A project lead will spot in thirty seconds the sentence that is technically true and practically misleading.

A workable rhythm

Scale comes from a steady process more than from volume on any one day. The shape that tends to hold up looks like this:

1. Choose pages from real questions: sales calls, support tickets, the searches that already reach you.
2. Gather the primary sources before anyone drafts.
3. Draft with AI help, working from those sources rather than the model's memory.
4. Edit for substance, then check every claim against its source.
5. Have someone who does the work review anything that describes it.
6. Publish with a review date, and put the recheck on a calendar.

The order matters more than the tools. Gathering sources first is what keeps the draft honest. Checking after editing is what catches the claims the editor introduced.

The common failures

Thin pages. A page that restates its title in three paragraphs and links to the contact form. Readers leave, and search engines have no reason to show it.

Invented statistics. "Studies show 78% of buyers..." with no study behind it. AI models produce these easily. A fact-check catches them only if someone looks.

Near-duplicates. Twenty pages that differ only in a city name or a product name. Bing's webmaster team has written that duplicate pages dilute signals, that AI systems group near-duplicates and pick one to represent the set, and that you should only keep separate pages when the intent really changes.

Confident errors. A page states that a product does something it doesn't. The reader finds out on the sales call, which is the worst possible time.

Signs the process has slipped

• Pages go live faster than anyone could have read them.
• Sources are other blogs, or nothing at all.
• The same paragraph appears, lightly reworded, on dozens of pages.
• Nobody can say when a page was last checked.
• The sales team has stopped sending prospects to the site because they don't trust it.

Any one of these is worth stopping for. Fixing the process is cheaper than cleaning up a few hundred pages later.

Why fewer, better pages win

A page per question your buyers really ask is a good plan. A page per keyword variation is not.

Good pages compound. They earn links, get cited in AI answers and keep converting for years. Template pages tend to sit unread, and a large pile of them can make the rest of the site look less trustworthy.

A useful rule: if you can't name the real person whose question a page answers, and what that page tells them that no other page on your site does, don't publish it yet.

What most guides miss

Disclosure is a choice worth making on purpose. Google's helpful content guidance suggests explaining how content was made where readers would reasonably want to know, including substantial use of automation. Decide your policy before someone asks.

Review dates need a calendar, not a field. Adding "Last reviewed" to a page is easy. Changing the date without rechecking is worse than having no date at all. Google lists changing dates to look fresh, when nothing meaningful changed, as a warning sign.

The best source is often your own team. The facts nobody else can publish are the ones in your project notes and your sales calls. AI can't invent those, which is exactly why they are worth publishing.

Removing pages is part of the job. Merge or redirect pages that no longer earn their place. A smaller, sharper site is easier to keep accurate, and every page you keep is a page someone has to recheck. Count that cost before you publish, not after.

How we do it

We start with a free intro call about what you want to publish and who it's for. Then we work alongside your team: AI for research and drafts, every claim checked against a primary source, a human editor on every page, and a review from someone who does the work. Our own site is produced this way. See websites and content for more.

Sources

1. Google Search Central: Spam policies for Google web search
2. Google Search Central: Guidance on using generative AI content on your website
3. Google Search Central: Creating helpful, reliable, people-first content
4. Google Search Central Blog: Google Search's guidance about AI-generated content (2023)
5. Bing Webmaster Blog: Does duplicate content hurt SEO and AI search visibility?
6. Bing Search Blog: Evolving role of the index