Every few months a small business owner asks me some version of the same question: "Will Google penalize my site if I use ChatGPT to write content?"

Short answer: no.

Longer answer: Google will absolutely penalize you for what most people do with ChatGPT, which is a different problem, and one worth understanding before you publish another page.

What Google said

In February 2023, Google published guidance on AI-generated content. The line people keep missing: "Our focus on the quality of content, rather than how content is produced, is a useful guide."

Then in March 2024, Google added a spam policy called scaled content abuse: generating lots of pages primarily to manipulate rankings rather than help people. Note the word "primarily." The method isn't the violation. The intent and the output quality are.

The rules didn't change when AI showed up. Thin content ranked badly in 2015 and it ranks badly now. What changed is the speed: a page that used to take an afternoon now takes ninety seconds, so more people are stepping on the same rake.

Google's own March 2024 announcement said the update was expected to cut low-quality, unoriginal content in search results by 40%. Those sites weren't hit for using AI. They were hit for publishing 400 pages of nothing.

Where AI genuinely earns its keep

A few places where it saves local businesses real time, with no ranking risk.

Topic research. Ask Claude or ChatGPT: "What questions do homeowners in Phoenix search for about AC repair in July?" You'll get 20 topics in thirty seconds. Some are obvious, some you wouldn't have thought of, a few are wrong. That's fine. Treat every item as a lead to verify. Cross-check the good ones against Google's autocomplete and the "People also ask" box before you commit to writing anything.

Meta tags. Title tags and meta descriptions are tedious and there are a lot of them. AI is good at this because the constraints are mechanical: aim for roughly 60 characters on titles and 155 on descriptions, keyword near the front, sounds like a human wrote it. Those are rules of thumb, not limits. Google truncates on pixel width, and the cutoff moves. It also rewrites meta descriptions more often than not, which is exactly why you shouldn't spend an afternoon hand-polishing them. Give it your page content and your target keyword, ask for five options, pick the one that doesn't sound like a robot. Ten minutes instead of an afternoon.

Page structure. If your service pages are a wall of text with no headings, AI will fix that in one pass. Ask it to organize your existing copy into a clear heading hierarchy with an FAQ section at the bottom. The words stay yours; only the arrangement changes. That's a formatting problem, and it's very good at formatting problems.

Voice notes into drafts. Talk through a job on the drive back from it, then hand the transcript over and ask for a clean draft. The details are yours because you dictated them from memory. All the model does is take out the "ums" and put the sentences in order. This is the highest-value use on the list, and almost nobody does it. If your drafts keep coming back generic, the problem is usually the ask, and the three-step prompt framework fixes most of it.

Where it goes wrong

The city page trap gets people every time. You serve Scottsdale, Tempe, Mesa, Gilbert, and Chandler, so you generate five pages with the same 600 words and the city name swapped. It feels like a shortcut. It's the single most reliable way to get your pages ignored.

Google has been detecting near-duplicate content since long before AI existed. Those pages either get folded into one another in the index or they just sit there earning nothing. Worse, they dilute the authority of the page that was earning something.

If you can't say something genuinely different about your Mesa service area than your Gilbert one, you don't need two pages. You need one good service page and a properly filled-out Google Business Profile. But if those five pages are already indexed, don't just delete four of them. Open Search Console first and see which ones are earning impressions, because one of them may be quietly doing work. Then 301-redirect the pages you cut to the service page that survives, so whatever links and history they picked up land somewhere instead of on a 404.

Publishing without reading is the next one. AI models make things up with total confidence. I've seen a generated plumbing page confidently list "trenchless sewer replacement" as a service the company didn't offer. The customer who calls about it and gets told no doesn't think "interesting AI hallucination." They think you're unreliable, and they call the next guy. That is the whole failure mode, and it never shows up in your analytics.

Then there's keyword density theater. Ask AI to "optimize for Phoenix HVAC repair" and it will cheerfully write you a paragraph containing "Phoenix HVAC repair" six times. Real people read one sentence of that and leave. A page written for the crawler gets skimmed and abandoned, and you lose the call either way. Writing for the crawler at the expense of the human is a losing trade, and it has been for a decade.

The split that works

Split the job at the line where knowledge starts.

AI handles the parts that don't require knowing anything about your business: structure, first-draft prose, headings, meta tags, FAQ scaffolding, cleaning up your grammar. Fast, cheap, low-stakes.

You handle the parts only you know:

  • Which neighborhoods you actually cover, and where you draw the line. "We service south Scottsdale but not past the 101" is more useful than any generated sentence about "serving the greater Phoenix area."
  • Real jobs. The 1974 house in Arcadia where the ductwork was in the crawlspace. The strip-mall tenant improvement that needed three permits.
  • Specific brands and parts. If you install Trane and Mitsubishi mini-splits, say so. People search by brand.
  • Actual pricing signals. Even a range beats silence.
  • Your real constraints. Same-day service until 4pm. No weekend calls. Say it.

That's the part no model can generate, because it isn't on the internet. It's in your head and your job history.

A practical rule I give clients:

If a competitor could publish your page verbatim and it would still be true, the page is worthless.

Rewrite it until that's no longer possible.

One more thing about volume

A lot of AI SEO advice pushes publishing cadence. Four posts a week, thirty pages a quarter.

For a local service business, that's usually wrong. Across the local service sites we manage, the homepage plus the top two or three service pages carry most of the organic sessions, with the Google Business Profile taking a large share of the rest. Your numbers will differ, so check your own GA4 before you take my word for it. But as a rule of thumb, twelve solid pages beat eighty mediocre ones, and twelve is achievable without AI writing anything you haven't read carefully.

Use AI to make those twelve pages faster to produce and better organized. Don't use it as a reason to have eighty.

Once those pages are solid, the next gains are off the page entirely: reviews, citations, and proximity. We went through those in Beyond the Map Pack, and none of it is work AI can do for you.

Get a second opinion on your pages

If you've already published AI-assisted content and you're not sure whether it's helping or quietly dragging you down, we'll take a look. Send us your URL through the form and we'll come back with which pages are underperforming, why, and what fixing them would involve. Usually a couple of days, and it's a written summary, not a pitch deck or a retainer conversation.

Request a local SEO review at marshland.software/contact.