The short answer
The best AI prompts for SEO use AI to research intent, build briefs, edit and check facts, while people add first-hand experience. Google Search Central's guidance, updated on 1 October 2026, says generative AI is useful for researching and structuring content, but generating many pages without adding value can breach its scaled content abuse policy. These 15 prompts cover keyword intent, briefs, editing, metadata and audits in ChatGPT, Claude, Gemini or Perplexity.
| Prompts on this page | 15, in 5 groups |
|---|---|
| Google's rule | AI content is fine if it adds value; mass pages without value breach the scaled content abuse policy |
| Special AI Overview tricks | None needed, according to Google Search Central |
| Best for research | Perplexity, because every answer shows sources |
| Best for editing | Claude, for careful work on long drafts |
| Best for data | ChatGPT, for Search Console spreadsheets |
What Google says about AI content and SEO in 2026
Before you use any of these prompts, know the rules they are built around. Google Search Central's guidance on using generative AI content, last updated on 1 October 2026, says AI is particularly useful when researching a topic and adding structure to original content. The same page warns that using AI tools to generate many pages without adding value for users may violate Google's spam policy on scaled content abuse. That spam policy, updated on 28 August 2026, lists mass-producing pages with generative AI tools that add no value for users among its examples, next to scraping and stitching together other sites' content.
Google's helpful content guidance, updated on 5 October 2026, adds the standard your pages are judged against. It asks whether content offers original information, reporting, research or analysis, whether it shows first-hand experience and whether it contains easily verified factual errors. Of experience, expertise, authoritativeness and trustworthiness, Google says trust matters most. It also flags warning signs such as changing the date on pages to make them look fresh when the content has not substantially changed.
Three practical rules follow, and every prompt below respects them:
- AI does the research, structure and checking; people add the experience. The prompts ask the model to find gaps, outline and edit, and they ask you to supply the first-hand material.
- Every fact gets checked. Google's AI guidance reminds publishers that generative models predict likely words rather than look facts up, so they can produce confident errors. Fact-check before you publish.
- Metadata counts as content. Google asks you to review AI-written titles, meta descriptions, structured data and image alt text, because they can show up in search results.
On AI Overviews and AI Mode, Google's AI features page says there are no additional requirements and no special optimisations needed: a page has to be indexed and eligible to show with a snippet, and normal SEO practice applies. You do not need special AI text files or extra schema for them. Traffic from these features is counted in the Search Console Performance report under the Web search type, which matters for prompt 14.
Keyword research and search intent prompts
The first three prompts turn a raw keyword list into decisions about pages. Run the intent map in any chat assistant. Run "Find the questions real people ask" in a tool that searches the web and shows sources, because you need to click through and confirm each question exists. Perplexity lists numbered sources under every answer, which makes it the quickest for this step. "Spot what the top results miss" works best when you paste the headings yourself, so the model compares real pages instead of guessing what ranks.
Content brief and outline prompts
A good brief is where AI saves the most time without touching the quality bar. The brief prompt makes the model separate what it can draft from what only your team can supply. The interview prompt is the most underrated one on this page: twenty minutes with your own expert gives you the original detail Google's guidance keeps asking for, and AI is good at writing questions that get past the obvious answers. Use the comparison outline for "A vs B" pages, where readers want a verdict and the honest catch more than a feature list.
Prompts to edit drafts into people-first content
These three prompts are the safety net. "Edit a draft into people-first content" asks the model to judge each section against Google's own self-assessment questions, which is a stricter test than "make this better". The fact-check prompt deliberately tells the model not to rule on truth from memory; it only lists claims and the source type to check, which plays to what models do well. The refresh prompt uses your real query data, so updates answer what searchers are asking now rather than moving the date.
Title, meta description and structured data prompts
Search snippets are where AI-written metadata most often goes wrong, usually by promising more than the page delivers. The title prompt forces accuracy and one natural use of the keyword. The structured data prompt tells the model to use only visible information and to leave out unknown fields; Google's AI guidance asks publishers to make sure structured data follows its policies and eligibility rules, and its AI features page says markup should match the visible text. Validate the output before it goes live.
Audit and internal linking prompts
For the Search Console prompt, export the Performance report as a spreadsheet and upload it. ChatGPT's data analysis and Claude's code execution both handle files like this; check the numbers in the output against the export for one or two rows before you trust the rest. The scaled content prompt is the one to run before any programmatic project, such as one page per city. If the honest answer is that most pages would only swap a place name, merge them.
Which AI tool works best for SEO
| Job | Best tool | Why |
|---|---|---|
| Question and competitor research | Perplexity | Every answer lists numbered sources you can open and check |
| Briefs, outlines and editing long drafts | Claude | Strong at careful writing and judgement across long documents |
| Search Console and keyword spreadsheets | ChatGPT | File and spreadsheet analysis on every plan, with limits on Free |
| Long research reports and Google Docs work | Gemini | Deep Research on the free tier and work inside Docs and Drive |
Watch your data. On consumer plans your chats can be used for training by default: in ChatGPT the switch is Settings, Data controls, Improve the model for everyone; in Claude it is Settings, Privacy, Help Improve our AI models; in Perplexity it is Account settings, Preferences, AI data retention. Turn these off before pasting client Search Console data, or use a business plan. For a fuller view of tools, read the best AI for SEO and the best AI for research.
How to adapt these SEO prompts
- Fill every placeholder with specifics. "Small accounting firms in Leeds" gives a far better intent map than "businesses".
- Paste, do not describe. Real headings, real queries and the real draft beat a summary every time.
- Keep the constraints. Lines such as "do not add new facts" and "leave out any property I have not given you" are what stop invented details. If you shorten a prompt, keep those.
- Chain them. Run 1, 3, 4, 5, then write, then 7, 8, 10 and 13. The output of each step becomes the input of the next.
- Say how the page was made when readers would wonder. Google suggests explaining how automation was used where readers would reasonably ask, such as a note on a data-heavy page.
New to writing prompts? Start with how to write a good prompt and how to reduce AI hallucinations, or try the Perplexity lessons.
Prompts to copy
Map search intent for a keyword list
You are an SEO strategist. Below is a keyword list for [site or business], which offers [product or service] to [audience] in [country]. For each keyword, label the likely search intent (informational, commercial, transactional or navigational), say what kind of page would satisfy it (guide, comparison, product page, tool, local page) and group keywords that one page could answer together. Return a table with the columns keyword, intent, best page type and cluster name. Flag any keyword where the intent is unclear and tell me what to check in the live results. Keywords: [paste list]
You get: A clustered keyword table that shows which page each group needs
Find the questions real people ask
Search the web for the questions people ask about [topic] in [year]. Look at forums, Q&A sites and community discussions, not SEO blogs. List 20 questions in the words people actually use, group them by stage (just curious, comparing options, ready to buy) and cite the source for each one. Then mark every question that our page about [URL or page topic] does not answer yet.
You get: Cited questions in the reader's own words, plus the gaps in your page
Spot what the top results miss
Here are the headings and main points from the current top [5] results for "[keyword]": [paste]. Our audience is [audience] and our first-hand experience is [what you have done, tested, measured or seen with clients]. List: 1) what every result covers, which is the minimum we must cover, 2) what none of them covers well, 3) three angles where our experience gives us something original to say. Do not suggest padding or repeating the keyword.
You get: The table stakes for the topic plus the original angles only you can write
Write a one-page content brief
Create a content brief for a page targeting "[primary keyword]" for [audience]. Include: the reader's main question in one sentence, the answer they need in the first 100 words, an H2 and H3 outline, the facts and figures we must verify (with the type of source to use for each), the first-hand material we need from [name or team] such as photos, results and examples, internal links to [list of our pages] and what to leave out. Keep it to one page.
You get: A brief that a writer or a subject expert can work from
Interview your subject expert
I am writing about [topic] for [audience]. Our subject expert is [role, for example our head of customer support]. Write 12 interview questions that draw out first-hand experience: real cases, numbers they have seen, mistakes, what they would do differently and opinions that disagree with common advice. Order them from easy to deep, and add one follow-up question after each.
You get: An interview script that gives your article material nobody else has
Outline a comparison page
Outline a comparison page for "[A] vs [B]" for [audience] who need to choose for [use case]. Start with a verdict, then one H2 per decision factor that matters to this audience ([factors]), a comparison table, who should pick which and the honest catch for each option. Mark every claim that needs a source with [VERIFY] and name the official page I should check.
You get: A comparison outline with every fact flagged for checking
Edit a draft into people-first content
Act as an editor using Google's helpful content self-assessment questions. Read the draft below and, for each section, say whether it adds original information, analysis or experience, or only restates what is easy to find elsewhere. Then cut filler and repetition, move the direct answer to the top, mark sentences that sound generic and list the claims that need a source. Do not add new facts. Draft: [paste]
You get: A tighter draft plus a list of weak sections and claims to check
Build a fact-check list before publishing
List every factual claim in the text below as a numbered table with the columns claim, type (number, date, name, product detail, quote, other), risk if wrong (high, medium, low) and the best primary source to confirm it. Do not judge from memory whether a claim is true; I will check each one myself. Text: [paste]
You get: A verification checklist, ordered by risk, to work through before you hit publish
Plan an honest refresh of an old page
Here is a page we published on [date] and its Search Console data for the last [3] months (queries, clicks, impressions, average position): [paste]. Suggest substantial updates only: outdated facts to recheck, questions from the queries that the page does not answer, sections to cut and new first-hand material to add. Do not suggest changing the published date unless the content changes meaningfully.
You get: An update plan built on real queries, not a cosmetic date change
Title tags and meta descriptions
Write 5 title tag options under 60 characters and 3 meta descriptions of 140 to 160 characters for the page below. Each must describe what the page actually delivers, use "[primary keyword]" naturally once and avoid clickbait or any promise the page does not keep. Page: [paste the page or a short description of it]
You get: Accurate titles and descriptions to choose from and test
Structured data that matches the page
Write JSON-LD for this page using the schema.org type [Article, Product, LocalBusiness, Organization or another type]. Use only information that is visible on the page below. Leave out any property I have not given you rather than guessing, and list the properties that are missing so I can decide whether to add them to the page. Page content: [paste]
You get: Structured data with no invented fields, ready to validate
Alt text that describes, not stuffs
Write alt text for each image below, which appear on a page about [topic]. Describe what each image shows and why it matters on this page in one short sentence, without stuffing keywords. If an image is purely decorative, say so instead. Images: [describe each image or attach them]
You get: Useful alt text for every image, and a list of decorative ones
Plan internal links for a new page
Here is a list of our pages with URL, title and main topic: [paste]. For the new page "[title]", suggest 5 to 8 existing pages it should link to and 5 pages that should link to it. For each link give natural anchor text and one sentence on why the link helps the reader at that point.
You get: An internal linking plan with anchor text and a reason for each link
Find opportunities in a Search Console export
Attached is a Search Console performance export of queries and pages for the last [3 or 6] months, with the previous period for comparison. Find: pages with many impressions but a low click-through rate, queries in average positions 5 to 15 that a page could answer better, and pages losing clicks against the previous period. Show a short table for each and give the likely cause in one line. Tell me what the data cannot show.
You get: Three prioritised lists of pages to fix, with the numbers behind them
Check a page template for scaled content risk
We plan to publish [number] pages from the template below, one per [city, product or keyword]. Review it against Google's spam policy on scaled content abuse: which parts would be identical or nearly identical across pages, what unique and useful information each page must have to deserve to exist, and which planned pages we should merge or drop. Template: [paste]
You get: An honest verdict on a programmatic page plan before it costs you rankings