Two things happened at once. Language models became the fastest keyword-clustering, intent-labeling, brief-writing, regex-generating assistant you have ever had. And Google started answering queries directly with AI Overviews, then AI Mode, so a page that ranks first can lose the click anyway. The job now is to do the mechanical work faster and to become the source the answer engines cite.
Google's position has been stable since early 2023: AI-generated content is not against its guidelines by itself; content produced primarily to manipulate rankings is, however it was made. The March 2024 spam-policy update named scaled content abuse and site reputation abuse, and the sites that got hit were mostly ones publishing thousands of thin generated pages. Experience, expertise, authority, and trust still decide, and thin output fails them at scale.
This page is about the workflow: what to hand a model, what to check before it ships, and how to think about visibility in AI answers without abandoning the fundamentals that still drive most traffic.
Quick wins this week
- Export 500 keywords from your rank tracker, paste them into Claude or ChatGPT, and ask for clusters by topic and intent with a suggested page for each; sanity-check the top ten clusters against live SERPs.
- Paste the H2s of the top five ranking pages for a target query and ask what every page covers, what only one covers, and what none do. That gap list is your brief.
- Ask for a regex that isolates question queries in Google Search Console, paste it into the filter, and find the FAQ content you rank for but never wrote.
- Generate JSON-LD for a page, validate it in Google's Rich Results Test, and delete any property the page does not visibly support.
What AI can do for seo specialists, task by task
Keyword clustering and intent classification
Paste keywords with your tool's volumes and current ranking URLs and ask for clusters a single page could satisfy, with intent and page type. The model clusters by meaning, not by SERP overlap, so verify the top clusters against live results before you commit a page to them. Never ask a model for search volume or difficulty; it does not have the data and will invent numbers.
SERP analysis and content briefs
Copy the titles, headings, People Also Ask questions, and any AI Overview text from the live SERP into the prompt, and ask for the table-stakes subtopics, the gaps, and the first-hand elements the page needs. A brief written from the model's memory describes the SERP of two years ago. Add your own data, screenshots, and expert quotes as requirements.
Drafting and refreshing content with real expertise
Draft from an expert's notes or interview, with the brief as structure, and ask for NEED markers where the source material is thin. Add first-hand experience, named authorship, and original data; that is what separates a page from the generated consensus. Fact-check every claim, and never publish dozens of pages at once from templates; that is the pattern Google's scaled content abuse policy describes.
Titles, meta descriptions, and internal links at scale
Export URL, current title, H1, and target keyword to a sheet and process it in batches, with character limits and a house pattern in the prompt. Ask for internal link suggestions from a crawl export with varied anchor text. Review every row; the model will truncate brand names and invent product claims to fit a limit.
Structured data generation
Ask for JSON-LD built only from content visible on the page, validate it in the Rich Results Test, and remove any property the page does not show. Google's structured data guidelines treat markup that misrepresents page content as spam, and fake review or pricing markup is a manual-action risk. The model will add ratings and dates that are not on the page if you let it.
Technical audits, regex, and log analysis
Paste crawl exports, redirect maps, or log samples and ask for the analysis and the script that produced it, so you can rerun and inspect it. Claude Code or ChatGPT will write the Python for log-file analysis and the regex for Search Console filters. Verify Googlebot by reverse DNS or Google's published IP ranges before you trust a crawl-frequency chart.
Visibility in AI Overviews, AI Mode, and chat search
Run your key queries in Google, ChatGPT search, and Perplexity monthly and record whether an AI answer appears and which sources it cites. Structure pages so an answer can be lifted from them: a direct answer near the top, clear entities, and supporting detail with sources. Be skeptical of anyone selling 'GEO ranking factors'; the engines have published little, and citation patterns shift monthly.
Prompts for seo specialists
Replace the bracketed placeholders, paste into any chat assistant, and iterate on the result.
Cluster keywords by intent
You are an SEO strategist. Below is a list of keywords with my tool's monthly volume and current ranking URL, if any. Cluster them into topics a single page could satisfy, using search intent (informational, commercial, transactional, navigational) and likely SERP format. For each cluster give a name, the primary keyword, secondary keywords, intent, the page type that fits, and whether an existing URL should own it or a new page is needed. Do not invent or adjust volumes. Flag clusters where you are unsure the SERPs overlap so I can check.
[PASTE KEYWORD, VOLUME, RANKING URL ROWS]Tip: Paste in batches of a few hundred; past that the clusters get vague.
Content brief from the live SERP
Act as a senior content strategist. Target query: [QUERY]. Audience: [AUDIENCE]. Below are the titles, headings, and word counts of the top results, the People Also Ask questions, and any AI Overview text I copied. Build a brief: the searcher's job to be done; subtopics every top result covers; subtopics only one or two cover; what none cover that an expert would; questions to answer in short, quotable paragraphs; required first-hand elements (examples, screenshots, data we own); three title options under 60 characters; and internal links from [PASTE RELEVANT URLS]. Do not write the article. [PASTE SERP DATA]
Tip: The 'what none cover' section is where the ranking opportunity usually is.
Schema markup from visible page content
You are a structured-data engineer. Generate valid JSON-LD for [SCHEMA TYPE, E.G. PRODUCT, FAQPAGE, ARTICLE] using only information that appears visibly on the page content below. Follow Google's structured data guidelines: add no properties (ratings, prices, dates, authors) that are not on the page and mark up nothing hidden from users. After the code, list each property with the exact page text it came from so I can verify, and note anything the page should display before I add a property. Page content: [PASTE VISIBLE PAGE TEXT]
Tip: Validate in the Rich Results Test, then confirm the page shows users the same content.
Search Console regex and query mining
Act as a Google Search Console power user. I want to find [WHAT YOU WANT, E.G. QUESTION QUERIES WITH IMPRESSIONS BUT LOW CLICKS ON PAGES UNDER /BLOG/]. Write the regex filters for the Query and Page dimensions that GSC's custom regex option accepts (RE2 syntax), explain each pattern in one line, and give three variations: broader, narrower, and excluding brand terms for [BRAND]. Then describe the comparison to run (date range, device, country) to spot pages losing clicks to AI Overviews, and what pattern would suggest that rather than a ranking drop.
Tip: Rising impressions with flat or falling clicks at a stable position is the pattern to watch.
Rater-style quality review
You are a Google Search Quality Rater applying the published guidelines. Assess the page below for the query [QUERY]. With reasons, rate whether it shows first-hand experience, demonstrates expertise, comes from an author and site trustworthy for this topic, fully satisfies the intent, and contains anything that reads as generated filler or copied consensus. List the specific paragraphs a rater would consider unhelpful and what original element (data, example, screenshot, process detail, opinion) would fix each. Be harsh; a polite review is useless to me. Author and site context: [DESCRIBE] Page: [PASTE PAGE TEXT]
Tip: Run it on your ten highest-traffic pages first; those are the ones a core update would hit.
Want a prompt for something else? Use the Prompt Builder or browse marketing and content prompts.
Skills to build
Knowing what the model does not know
Why: A language model has no search volumes, no live SERPs, and no ranking data, and it will produce all three on request.
How: Keep your rank tracker and Search Console as the only source of numbers, and give the model data instead of asking it for data.
Grounding briefs in the real SERP
Why: The model's idea of what ranks is a memory of the web from its training cutoff; the SERP you are competing on is today's.
How: Copy live results, PAA questions, and AI Overview text into every brief prompt, and rerun before publishing if more than a month has passed.
Adding first-hand experience
Why: Experience is the one quality signal a generated draft cannot supply, and it is what raters and AI answers reward.
How: Build an interview or data step into every content workflow: the expert's notes, your own screenshots, a test you ran, a number you own.
Reading Google's documentation, not summaries of it
Why: Half the SEO advice about AI content is paraphrase of paraphrase, and the primary sources are short and public.
How: Read Google Search Central's AI content guidance, the spam policies, and the structured data guidelines once a quarter, and note the dates.
Light scripting with an AI pair
Why: Technical SEO at scale (logs, crawls, redirects, migrations) is scripting work, and a model makes Python and regex accessible without a developer.
How: Use Claude Code or ChatGPT to write scripts you read and run yourself, starting with a log parser and a redirect-map checker.
Tools worth knowing
Claude
A careful writing and analysis assistant that shines on long documents.
ChatGPT
The general-purpose AI assistant most of your coworkers already use.
Perplexity
An answer engine that cites its sources, built for research rather than chat.
Gemini
Google's assistant, strongest when your work already lives in Google Workspace.
Julius AI
Chat with your spreadsheets and data files and get charts, stats, and answers back
Claude Code
Anthropic's agentic coding tool that works in your terminal and IDE, on your real codebase
Cautions for seo specialists
Google says AI-generated content is not against its guidelines by itself, and that using automation to produce content primarily to manipulate rankings is spam. Its scaled content abuse policy targets mass-produced pages that add little value however they were made, and site reputation abuse targets third-party content published on a host site to exploit its rankings. Consequences range from core-update losses to manual actions. Publish what a person would want to read, at a pace a person could review.
Models fabricate search volumes, difficulty scores, click-through curves, and confident lists of 'ranking factors', and now 'GEO factors' for AI answers. Use only numbers from your tools, treat any factor claim without a primary source as opinion, and test before you build a strategy on it.
Markup that does not match visible content, fake review ratings, or product schema that disagrees with the page can draw a manual action and lose rich results for the whole site. Generate from page content only, validate, and audit existing schema before you add more.
Search Console exports, revenue by page, unreleased product names, and migration plans are confidential: never paste confidential data into a consumer AI tool unless the client or your organization has approved it, and strip revenue columns before analysis. Health, finance, legal, and safety pages are held to a higher standard, and a generated draft will include fabricated statistics; route those pages through a qualified reviewer and cite primary sources.
Your 30-day plan
- Week 1: Read Google's AI content guidance, spam policies, and structured data guidelines. Set data rules: business accounts, no revenue or unreleased plans in consumer tools.
- Week 2: Cluster one keyword set and write SERP-grounded briefs for three pages, with a first-hand element required in each.
- Week 3: Generate and validate schema for one template, build a Search Console regex set, and have the model write one log-analysis script you run yourself.
- Week 3: Run the rater-style review on your ten highest-traffic pages and fix the honest findings.
- Week 4: Baseline AI visibility: 30 key queries across Google, ChatGPT search, and Perplexity, noting whether an AI answer appears and who is cited. Build the reporting template and schedule a monthly rerun.
Frequently asked questions
Does Google penalize AI-generated content?
Will AI replace SEO specialists?
How do I rank in Google AI Overviews and ChatGPT search?
Can ChatGPT do keyword research?
Terms used on this page
Related roles
- AI for MarketersDraft campaigns faster, mine customer feedback for messaging, and turn reporting into a ten-minute job, without drifting off-brand or into a CAN-SPAM problem.
- AI for Content WritersAI can research, outline, draft, and edit alongside you. The writers who thrive with it use it to reach the interesting part faster, keep their own voice, and are straight with clients about how they work.
- AI for Ecommerce ManagersYou run a catalog, a marketing calendar, a support queue, and a P&L, often with a team that is too small. AI can write listings from spec sheets, mine thousands of reviews, draft support replies, and forecast demand. It will also invent a product feature or a five-star review if you let it, and both are now expensive.
- AI for Social Media ManagersFeed the AI your voice guide and your best posts and it will draft a month of captions, cut a webinar into clips, and summarize what your audience is saying. You still decide what ships, and you still own the labels and the DMs.