AI is excellent at the grunt work of SEO and risky at the job most businesses hand it first, which is writing the content. Google’s October 2026 guidance made that line easier to see, and our AI search optimization team now splits SEO work along it.
Where AI earns its place in SEO
The SEO tasks that suit AI involve large amounts of data and produce output a specialist can check in minutes.
Keyword research is the clearest example. Give a model a few thousand keywords from Semrush and it will cluster them by topic and sort them by search intent far faster than a person with a spreadsheet. You still review the groups, but you start from a draft instead of a blank sheet.
Site audits are another. On a 20,000-page site, Semrush pulled the crawl, ranking and backlink data, and Claude grouped the issues by pattern and ranked the fixes by impact and effort. The full project is in our enterprise SEO audit case study. Competitive gap analysis works the same way: compare your keyword set with your competitors’, find the topics they rank for and you don’t, and turn the gaps into a content calendar. That one is written up as a competitive gap analysis case study, and both projects appear in Where AI earns its keep in marketing, and where it costs you.
Then there is the small, repetitive copy: meta descriptions, image alt text and schema markup. AI drafts these well. Check them anyway. Google’s updated guidance names titles, alt text and structured data among the places where generated errors creep in, and a meta description that promises something the page doesn’t deliver costs you clicks.
Where it hurts: long-form content at scale
The CMO Survey found that 73.9% of companies use AI in marketing for content creation. A lot of that content is written to rank in search, and that is the use I trust least.
Google’s spam policies define scaled content abuse as generating many pages mainly to manipulate rankings instead of helping users, “no matter how it’s created.” The first example on the list is using generative AI tools to generate many pages without adding value for users.
On October 1, 2026, Google went further in its guidance on generative AI content. It explains that models predict words instead of retrieving facts, warns that output can contain hallucinations, and calls manual fact-checking of all AI-generated content critical before publishing. Material added from the quality rater guidelines says that producing large amounts of AI text without manual oversight counts as little to no effort. The edit landed a week after Google’s September 2026 spam update began rolling out, though Google has not linked the two.
The pattern we see on sites that lean on bulk AI articles is familiar: a climb in traffic for a few months, then a sharp drop, then a long, slow recovery once the site changes course. The articles are rarely spam in the obvious sense. They are just interchangeable with every other article on the topic, and they add nothing a searcher can’t find elsewhere. AI is the average of everything explains why that sameness happens.
A workflow for AI-assisted content that can rank
Google is not against AI content. It rewards effort, original information, real expertise and accuracy, and those come from people. This is the order we work in:
- Start with what only you have. A client result, original data, a strong opinion, or first-hand experience with the problem. If you have none of those for a topic, think hard before writing about it.
- Use AI for structure. Ask for outlines, competing angles and the questions searchers ask, then pick the angle yourself.
- Draft with the facts in the prompt. Paste in your data and notes so the draft is built on them.
- Rewrite in your own voice. Put the specific details, names and examples back in. Cut anything that could appear on any competitor’s site.
- Check every number, quote, title, meta description and schema field against a source.
- Publish fewer pieces, and measure them by rankings, leads and revenue instead of word count.
Text from newer Claude models now carries an invisible watermark, and ChatGPT text in the EU is getting one too. Google has not said it uses those signals in ranking. How the watermarks work, and what they can and can’t prove, is explained in Your AI tool is predicting words. Plan your marketing around that.
The same habits carry over to AI-driven search results, where being cited depends on having something original to cite. That is where most of our search work now goes.
A note for ecommerce sites
Google’s October update also restated Merchant Center rules. AI-generated product images need IPTC metadata marking them as made by a trained algorithm (the DigitalSourceType value TrainedAlgorithmicMedia). AI-generated product titles and descriptions have to be submitted separately and labeled as AI-generated. If your product catalog was written or illustrated with AI, check your feed before Google does.