Using AI Writing Without Hurting Your Clients’ SEO

A client asks the question in one of two moods. Either they are worried (“will Google punish us if you use AI?”) or they are hopeful (“can you just publish ten of these a week?”). Both deserve the same boring answer, and it comes from Google’s own documentation rather than from SEO folklore.

The short version: Google does not rank a page lower because a model helped draft it. It does act against pages that exist to game search results, and a freelancer who delivers unchecked AI text at volume is closer to that line than most realise. This guide covers what the documents actually say, where the real risk sits, and a checklist to run before you hand a draft over.

What Google says, in plain terms

Google’s guidance on generative AI content, last updated 1 October 2026 at the time of reading, does not ban the practice. It describes AI as useful for researching a topic and for adding structure to original work. The conditions are that the result serves readers and still meets Search Essentials and the spam policies (Google Search Central, generative AI content, read 11 Oct 2026).

The line Google draws is about purpose and value, not about tools. Its spam policies define scaled content abuse as producing many pages mainly to manipulate rankings instead of helping people, and they list generative AI used to mass-produce low-value pages as one example of it, next to scraping and stitching together other sites’ text (Google Search Central, spam policies, read 11 Oct 2026). The same policy applies if a human typed every word of those pages.

Common claim	What the documents support	What it means for your delivery
“Google penalises AI text”	Not as such. Value to readers and spam policies decide.	Judge the page, not the tool used.
“Ten posts a week is fine if each reads well”	Volume without added value is the scaled-content pattern.	Agree a pace the client can actually fact-check.
“The model is accurate enough”	Google notes that models predict likely wording and can be wrong, so review before publishing.	Verify every number, name and claim.
“Only the body needs a check”	The guidance extends review to titles, meta descriptions, structured data and image alt text.	Check the small fields too.
“Hide that AI was used”	Google suggests explaining how content was made where readers would expect it.	Agree a disclosure line with the client.

Nothing in that table is a loophole or a trick, which is the point. The SEO field has long separated methods that follow search engine guidelines from methods built to deceive them, and the second kind tends to end in lost rankings once detected (Search engine optimization, Wikipedia, read 11 Oct 2026). AI drafting is a tool inside that old distinction, not a new category.

Where the actual risk sits for a freelancer

It is rarely one article. The danger is a pattern across a client’s site that you helped create. Three versions show up most often:

• Same skeleton, new keyword. Forty posts with identical headings, a swapped product name and a generic closing paragraph. Readers notice before search engines do.
• Confident errors. A wrong price, an invented statistic or a feature the product does not have. This costs trust, and it is entirely avoidable.
• Nothing only this client could say. No process, no example, no customer detail. If a competitor could publish the same page by changing the logo, it adds nothing.

If your own work leans on drafts from a model, the detection question matters less than the quality question. The guide to AI detectors for freelancers explains why those scores are a poor test, and why the better defence is a record of how you worked.

A pre-delivery checklist that takes about twenty minutes

Run this on every AI-assisted draft before it leaves your hands. It is deliberately short, because a checklist nobody finishes protects nobody.

1. Find the client-specific material. Add at least one thing the model could not know: an internal process, a customer question from their inbox, a figure from their own reports. If you cannot find any, ask the client for it.
2. Verify every checkable claim. Prices, dates, quotes, product features. Open the source and note the date you read it. Delete anything you cannot confirm.
3. Cut the filler. Remove the throat-clearing opener, the “in today’s fast-paced world” line and any paragraph that restates the heading. The editing guide for AI content walks through the pass line by line.
4. Vary the shape. If the last three posts you wrote for this client opened with a question and closed with a summary, change one of them.
5. Check the small fields. Title tag, meta description, image alt text and any structured data. These are the parts people paste straight from the model without reading.
6. Look at the whole batch. Open the client’s last ten posts together. If they read like one article repeated, slow down before adding an eleventh.

Setting the pace with a client

Speed is the real temptation. A model makes a first draft in seconds, so the client’s mental price per post drops and the requested volume climbs. Hold the line on the part that does not speed up: checking and adding original material. A reasonable position to put in writing is that you will deliver a number of posts the client can review and stand behind, and that each one gets the checklist above.

That framing also protects you. When a client later asks why traffic is flat, you can point to a process rather than defend a pile of pages. Pair it with the pricing approach in the comparison of AI for writers and designers, which covers charging for judgement instead of for word count.

Should you tell readers AI helped?

Google’s guidance suggests explaining how content was made, where readers would reasonably expect that information, rather than leaving it unsaid. For client work this is a decision for the client, not for you, so raise it early. A short editorial note (“drafted with AI assistance, fact-checked and edited by our team”) is enough for most business blogs. Where your contract or the client’s policy restricts AI use, follow it. For privacy and tool choice, the Claude vs ChatGPT comparison for writers covers what each plan does with your inputs.

Who this approach suits, and who should skip it

It suits you if you write for a handful of clients, can ask them for real examples, and treat the model as a fast first draft. You will spend more minutes per post than a copy-and-paste workflow and ship pages you would be comfortable defending.

Skip it if the brief is hundreds of near-identical pages at the lowest possible price. That is the volume-first pattern Google’s own policy describes, and no checklist will fix it. Decline it or reprice it.

If you are still choosing tools for drafting, the Jasper vs Copy.ai breakdown shows what each costs today. Browse more in the AI writing and content section.

One thing to do this week

Pick one client and read their last five posts in a row, as a stranger would. Note anything that repeats, anything unverifiable and anything that could appear on a competitor’s site unchanged. That list is your next month of improvements, and it is worth more than any additional draft.