Daniel Carter
I've been doing SEO for over 25 years. AI hasn't replaced any of it. What it has done is take work that used to eat hours and turn it into work that takes minutes. That's a real change, but it isn't magic. The output is only as good as the input, and the input is still you. SEO's are scrambling to take advantage of AI and automation - unfortunately, this also includes the SEO grifters/spammers and blackhatters.
AI has its place within SEO, to improve workflows, to automate mundane tasks and to help with data analysis and even for things such as content brief generation - it is NOT however a replacement for an SEO themselves - only those who have tried will understand why.
This is the working version of how I actually use AI for SEO right now. Not the theoretical version. Not the one written by people who've never run a real client account. The version where you feed it bad prompts and get bad answers, where it hallucinates a backlink that doesn't exist, where the content it writes sounds like every other AI page on the internet unless you do something about it.
Here's what works and what doesn't.
What AI Is Actually Good At in SEO
The honest answer: AI is good at summarizing, drafting, classifying, and pattern-matching. It's poor at strategy, taste, and original insight. The moment you treat it like a strategist, it lets you down. The moment you treat it like a smart intern who needs clear instructions, you get a lot done.
Tasks where AI saves real time:
• Generating outlines and briefs from existing pages
• Drafting meta titles and descriptions in bulk
• Writing schema markup
• Cleaning and analyzing GSC, Ahrefs, and Semrush exports
• Suggesting internal linking targets across hundreds of URLs
• Producing first drafts you'll then rewrite
• Explaining technical output like log files, server responses, and regex
Tasks where AI underperforms:
• Original keyword strategy. It will hand you the obvious terms everyone is already targeting - or, it'll fabricate search volumes
• Diagnosing why a page is dropping. It guesses from limited context. It can guess or make assumptions on data, but its rarely right on its own
• Writing copy that sounds like you wrote it. Not without serious prompting and editing.
• Predicting what Google will reward next. No model has insider info on the next core update.
Start from this baseline and you'll get more out of it than people who think it's about to replace your job.
Keyword Research and Topic Discovery
Most articles on this subject tell you to ask ChatGPT for a list of keywords. That's not keyword research. That's a brainstorm. Real keyword research uses search volume, click data, and SERP analysis. AI helps in the steps before and after the data work.
Where it helps:
Cluster Mapping
Take a list of 200 keywords from Ahrefs or Semrush. Paste them into Claude or ChatGPT with this prompt:
Group the following keywords into topical clusters based on user intent. For each cluster, name it, list the keywords inside it, and specify the primary search intent (informational, commercial, navigational, transactional). Output as a table.
You get a working cluster plan in under a minute. The same job in a spreadsheet takes two hours.
Question Discovery
AI is decent at producing the kinds of questions people ask around a topic. Prompt it like this:
I'm building a guide on [topic]. List 30 specific questions a buyer at the awareness, consideration, and decision stage would ask. Sort them by stage. Avoid generic questions. Be specific to [industry/audience].
Validate the output against AlsoAsked or PAA data, but it's a faster start than staring at an empty doc.
Long-Tail Variations
Give it a head term and you'll get a workable list of long-tail variations. Sense-check them against actual search volume. AI doesn't know what people are searching for. It knows what sounds like something someone might search for.
Briefs That Are Actually Useful
This is one of the highest-value uses of AI in SEO content production. A good brief saves writers hours. A bad brief produces filler.
Take the top 5 ranking URLs for your target keyword. Paste their headings or pull them programmatically with a tool like SEO Stack. Then prompt:
Below are the H1, H2, and H3 headings from the top 5 ranking pages for the keyword "[keyword]". Identify what every page covers, what most pages cover, and what only one page covers. Then identify topics no page covers that would add genuine value to a reader.
That last part is what separates a brief from a copy of what's already ranking. You don't want to produce another version of the same article. You want to produce the better version, which means knowing what's missing.
Then ask:
Based on this analysis, draft a brief with the following sections: target keyword, secondary keywords, recommended H1, suggested H2 structure, sections to expand beyond competitor coverage, and three sections that introduce new value not currently in the SERP.
That's a brief worth giving to a writer.
Writing First Drafts the Right Way
If you ask AI to write an article and publish it, you'll publish AI slop. Everyone can spot it. Google's quality systems can spot it. Your readers can spot it. Hitting the keyword doesn't help.
Use it as a draft engine, not a writing replacement. Here's the workflow:
1. Feed it the brief.
2. Give it the audience. Specific, not generic.
3. Give it your tone of voice as examples. Paste 500 to 800 words of your actual writing.
4. Ask it to draft section by section, not the whole article at once.
5. Rewrite every section by hand.
The rewrite is the part you can't skip. AI doesn't know what your customer's actual problem is, what your insider perspective is, or what stories you've seen play out. You add those in the rewrite.
Avoid asking it to "write in the style of [your name]." It can't, and the result will read like a parody. Give it samples and tell it to match the rhythm, sentence length, and vocabulary range. Then edit.
Meta Titles, Descriptions, and Alt Text at Scale
This is where AI earns its keep on large sites. A site with 500 product pages and no meta descriptions used to be a job. Now it's a job for a Saturday morning.
For meta descriptions, prompt:
For each of the following product titles, write a meta description that is 150 to 160 characters, includes the primary keyword once, contains a clear benefit, and ends with a call to action. Output as a CSV with two columns: product_title, meta_description.
You can run hundreds of products in batches. Same approach works for title tags. Same approach works for image alt text once you describe what's in each image.
The catch: spot-check the output. AI will sometimes pad descriptions with filler or repeat the same call to action across 200 products. Skim through. Fix the ones that miss.
Schema Markup Without the Headache
Schema is a slog if you write it by hand. AI handles it well because the format is structured and predictable.
Give it the URL, the page type, and the elements you want marked up:
Generate JSON-LD schema for the following article. Include Article, BreadcrumbList, and Author schema. Article details: [paste in title, author, publish date, modified date, image URL, description]. Output valid JSON-LD only, no commentary.
Validate it in Google's Rich Results Test before deployment. AI gets schema right most of the time but will occasionally invent a property that doesn't exist. The validator catches that.
On a site with 50 service pages, this turns a multi-day job into an afternoon.
Internal Linking at Scale
This is one of the underused angles I see most teams miss. Internal linking has a real ranking effect, and AI is good at finding link opportunities once you give it the data.
The method:
1. Export a list of all your URLs with their primary keywords or topics.
2. Paste it into Claude or ChatGPT in chunks of 50 to 100.
3. Ask it to identify three relevant link targets for each URL with reasoning.
The prompt:
Below is a list of URLs and their topics. For each URL, identify three other URLs from the list that would be relevant internal link targets, explaining why each makes sense based on topical proximity. Output as a table with columns: source URL, target URL, anchor text suggestion, reasoning.
You'll get a working internal linking map. Apply the obvious ones. Skip the suggestions that don't make sense. The reasoning column is what makes this useful, because it forces the model to justify each suggestion.
Technical SEO Tasks Most People Don't Use AI For
Most articles on this subject focus on content. Technical SEO is where AI is just as useful, sometimes more so.
Regex
GSC filters, redirect rules, log file parsing. If you don't write regex daily, you forget the syntax. Ask AI for it, then test it.
Write a regex pattern that matches all URLs containing /blog/ but excludes /blog/category/ and /blog/tag/. Test it against the following sample URLs and tell me which match and which don't.
Five minutes of work that used to be twenty.
Log File Analysis
Paste a chunk of your server log into Claude. Ask:
Below is a sample from my server log. Identify which user agents are crawling my site, what response codes they're hitting, and flag any patterns that suggest crawl budget issues, indexing blocks, or unusual bot activity.
You're not getting a full audit. You're getting a working summary that tells you where to look.
GSC Performance Investigation
Export a query report for a page that's lost rankings. Paste it in. Ask:
Below is a Google Search Console performance export for a single URL across two date ranges. Identify which queries lost the most clicks, which queries lost positions, which queries gained, and what the pattern suggests about content gaps or intent drift.
This is one of the most valuable AI use cases in SEO. It surfaces patterns you'd miss reading rows in a spreadsheet.
Optimizing for AI Search Engines (AEO and GEO)
This wasn't a topic two years ago. It is now. People are searching inside ChatGPT, Perplexity, and Google's AI Overviews, and the rules for showing up there overlap with traditional SEO but aren't identical.
What I've seen work:
• Direct, factual answers near the top of the page. AI engines pull from clear, declarative sentences. Bury your answer six paragraphs in and you're less likely to be cited.
• Structured data. FAQ, HowTo, and Article schema make it easier for LLMs to parse your content. Not a guarantee of citation, but a clear signal.
• Clean H2 and H3 structure. Models segment content by headings. If your headings describe what's in the section, you're easier to extract from.
• Original data and quotes. AI engines cite primary sources. Publish original research, surveys, or expert quotes and you're more likely to be referenced than a page rewriting what's already out there.
• Author bylines and authority signals. E-E-A-T applies in AI search too. Show who wrote it, why they're qualified, and what they've done.
What doesn't work: trying to game it. Stuffing your page with FAQs to trigger citations. Repeating the brand name 14 times. Writing for a model instead of a person. The signals are too sophisticated for that, and the bar gets higher every quarter.
A Quality Control Checklist for AI Content
Most AI content fails at the same points. Here's what to check before anything goes live.
1. Read it aloud. If a sentence sounds like it's been written by a robot, it has been. Cut it or rewrite it.
2. Check the facts. AI hallucinates. Names, statistics, dates, citations. Verify everything.
3. Strip the AI fingerprints. Phrases like "in today's fast-paced world," "navigating the complexities of," and any sentence that starts with "It's important to note that" need to go.
4. Add specific examples. AI writes in generalities. Replace at least one general claim per section with a specific example, number, or case.
5. Cut the throat-clearing. AI loves to introduce paragraphs with "When it comes to..." or "In the world of..." Delete them. Start with the point.
6. Make sure it has a perspective. Generic content explains things. Useful content takes a position. If your AI draft has no point of view, add one.
7. Compare it to the SERP. Does your draft offer something the top 10 don't? If no, keep going. You're not done.
8. Remove em-dashes. AI overuses them. Rewrite around them.
9. Check the meta description. AI will often write a description that doesn't match what's actually in the article. Verify.
10. Validate schema. Run any AI-generated JSON-LD through Google's Rich Results Test before publishing.
Skip these and you ship slop. Slop doesn't rank, doesn't earn links, and damages your topical authority over time.
Tools That Pull Weight in 2026
I won't list 30. Here's what I actually use:
• Claude or ChatGPT for everything content-related, briefs, regex, schema, and analysis. Both are strong. Claude tends to be better at long-form writing. ChatGPT has more integrations.
• Ahrefs and Semrush for keyword and backlink data. Both are essential. The AI inside them is decent but not a replacement.
• Surfer or Frase for content scoring against the SERP. Useful, not essential.
• Screaming Frog with the AI integration for technical work at scale. Schema generation, alt text, and meta descriptions across thousands of URLs.
• SEO Stack for connecting Google Search Console with AI analysis. Disclosure: I built it. The reason I built it is exactly the GSC analysis use case described above. Reading 10,000 rows of GSC data with your eyes is bad use of time.
• Google's Rich Results Test for schema validation. Free, official, mandatory.
What I don't use: most of the AI content writers. They produce content that ranks for a few weeks, then gets caught in the next quality update. Content quality has gone up, not down. Generic AI output is more visible to Google, not less.
A Final Thought
AI changes the speed of SEO work. It doesn't change the work itself. The principles are the same: understand the searcher, build something useful, signal trust, and earn the link. AI helps with the production. It doesn't help with the judgment.
The teams getting the most out of it are the ones who already knew what they were doing. They've added a tool to their stack. The teams getting the least are the ones who thought it would replace strategy. It won't.
If you're new to AI in SEO, pick two or three of the use cases above. Run them for a month. Build them into your workflow. Then add more. You'll find your own use cases that aren't on this list, because every SEO setup is different.
What you don't want to do is push a button and publish what comes out. That's the fast track to getting nothing back from your effort.