AI in SEO: what it does for each task, and where it stops
Short answer AI in SEO plays two roles. Inside Google, AI systems work out what a search means. On your side, AI tools speed up keyword ideas, drafts, titles, audits and link prospecting. They stop at facts and judgment: AI predicts words rather than looking them up, so a person checks every number, claim and page before it goes live.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callWhere does AI help in SEO, task by task?
AI saves the most time on jobs that need speed and volume: brainstorming, first drafts, sorting data, spotting patterns. It is weakest where the job needs true facts about your business or a decision about what to publish. Use it to start work and to check work, not to finish it.
| SEO task | What AI does well | Where it stops | What to check it against |
|---|---|---|---|
| Keyword research | Lists how customers might phrase a need, groups ideas by topic | It guesses search volumes; it has no live search data | Search Console's queries, a keyword tool's numbers |
| Search intent | Sorts searches into "wants info" vs "wants to hire" | Can't see what Google shows today for your city | The actual results page |
| Writing pages | Outlines, first drafts, rewrites for clarity | Invents facts, prices and credentials | Your own records, a person's read-through |
| Titles, descriptions, schema | Fast variations and code | Can produce wrong or rule-breaking markup | Google's Rich Results Test, a person |
| Page experience | Explains speed reports, suggests fixes | Can't measure what real visitors feel | Core Web Vitals field data |
| A/B tests and personalization | Drafts variants, reads test results | Doesn't know Google's testing rules | Google's A/B testing guidance |
| Links | Finds sites that fit, drafts outreach | Can't judge trust or relationships | You, before any email goes out |
| Technical audits | Crawls and sorts hundreds of issues | Ranks them by count, not by money | Which pages bring calls |
| Reporting | Summarizes trends, flags drops | Can't tell a seasonal dip from a real problem | Last year's same month |
AI plays two roles: inside Google, and in your tools
Google itself runs on AI. Its list of ranking systems names several. One of them:
"RankBrain is an AI system that helps us understand how words are related to concepts." (Google Search Central: A guide to Google Search ranking systems)
The same guide names BERT for how word combinations express meaning and intent, and neural matching for linking concepts in a search to concepts on a page. One correction to a common claim: MUM, Google's model that can also generate language, is "not currently used for general ranking in Search," only for specific uses such as featured snippet callouts.
What this means for you: Google already understands plain language. A page that answers "my AC blows warm air" in a customer's words does not need the phrase "AC repair" stuffed into every line. More in Google's AI ranking systems.
The second role is the AI tools you or your agency use. That is the rest of this page.
Keyword research: ideas from AI, numbers from real data
AI chat tools are good at producing the ways people describe a problem: "water heater leaking from bottom," "is a leaking water heater an emergency." Treat every number they give you as a guess. Google explains why:
"Keep in mind that generative models don't retrieve facts, but predict a likely sequence of words based on their training data." (Google Search Central: Google Search's guidance on using generative AI content)
Check the ideas against real searches. Search Console's Search performance report shows the actual queries that brought people to your site, by page and country. Then confirm volumes in a keyword tool. Step by step: AI keyword research and its limits.
Content: what Google allows from AI
Google does not ban AI writing. It bans using it to flood search. Its guidance says AI "can be particularly useful when researching a topic, and to add structure to original content." The line is in its spam policies:
"Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users." (Google Search Central: Spam policies)
Three rules follow from Google's pages:
- Fact-check everything. Google calls it "critical to manually factcheck and review all AI-generated content" before publishing, and says that covers titles, meta descriptions, structured data and image alt text too.
- No fake experts. Google's helpful content page calls AI-generated headshots, made-up names or false credentials "a form of deception."
- Add what only you know. Your prices, your service area, your photos, the questions customers ask on the phone. A page that only summarizes others is a warning sign on Google's own self-check list.
How to split the writing between you and a tool: AI content and SEO and AI tools and human creativity in SEO.
User experience: AI suggests, real visitors decide
AI can read a speed report and tell you which image to shrink. It can't tell you how your site feels on a customer's phone. Google measures that from real visits: Core Web Vitals "measure real-world user experience" for loading, responsiveness and visual stability. Google also says not to chase one or two numbers but to check for "an overall great page experience across many aspects." Benchmarks: Core Web Vitals scores.
Two AI-powered UX tactics come with Google rules:
- A/B testing. Fine, if you follow Google's testing page: don't show Googlebot different URLs than people, point test pages to the original with `rel="canonical"`, use a 302 (temporary) redirect, and "remove all elements of the test as soon as possible."
- Personalization. Changing a page per visitor is allowed, but Google's spam policies define cloaking as "presenting different content to users and search engines" to manipulate rankings. Whatever Google sees must match what visitors see. Details: dynamic content and SEO.
Links and technical SEO: AI finds, you decide
AI tools can sort thousands of sites and pick the few that fit your trade and town, then draft a pitch. The judgment call stays with you, because Google's spam policies list "buying or selling links for ranking purposes" as link spam. Any tool that sells "AI link packages" is selling that risk. See what link building tools can and can't do.
For technical SEO, AI crawlers list every broken link and missing tag. They tend to rank problems by how many pages have them. You rank them by which pages bring calls. How to run one: technical SEO audit.
What AI can't do for your SEO
- Know your business. Hours, licenses, warranties, real jobs. Only you have them.
- Decide what to publish. Google asks site owners to consider "Who, How, and Why" about their content. A tool can't answer why you made a page.
- Promise a result. Neither AI nor anyone else controls rankings. Google's AI answers follow the same rules: it says "there are no additional requirements to appear in AI Overviews or AI Mode." Next steps for that: adapt your SEO for AI search.
Example
A plumber with no time to write
AI drafts 4 service pages in an afternoon. The owner adds real prices, a photo of each job type and her license number, and corrects one wrong claim about permits. Four pages, about 3 hours of her time, instead of none.
A dental office tempted by volume
A tool offers 200 city pages for $500. Each page is the same text with a different town name. Google's spam policies call that scaled content abuse, and city pages built only to rank are doorway abuse. The office publishes 5 pages instead, one per town it actually serves, each with directions and that town's patients' common questions.
A landscaper reading an audit
An AI crawl flags 300 issues. Most are missing alt text on old blog photos. The owner fixes the 2 that matter first: a broken contact form and a slow quote page.
FAQ
Does Google penalize AI-written content?
Not for being AI-written. Google acts on pages made mainly to manipulate rankings, however they were made. Accurate, useful AI-assisted pages are fine.
Can AI do my SEO by itself?
No. It handles legwork: research, drafts, audits, reports. A person must fact-check, choose what to publish and approve outreach, as Google advises.
Is ChatGPT a good keyword research tool?
For ideas and phrasing, yes. For search volumes, no: it predicts words, not search data. Confirm with Search Console and a keyword tool.
Should I label pages as AI-written?
Google suggests it where readers would reasonably ask "How was this created?" For a normal service page you edited, a label is optional.
Different for you?
- Not sure your pages show up at all: Google Search Console for small business.
- Checking whether past SEO work paid off: is my SEO working?.
- Your site claims expertise you need to prove: E-E-A-T for small business.
- You'd rather hand the whole job off: SEO.