ChatGPT Prompts for SEO: 23 That Work, and How to Check Them

The ChatGPT prompts that work for SEO ask it to organize, compare and edit material you paste in. The ones that fail ask for facts it cannot know: search volumes, competitor traffic, rankings. Below are 23 copy-paste prompts in five groups, each with the check that catches its usual mistake.

Our stance: prompts speed up thinking you would do anyway; they do not replace data. The last group matters most for AEO (answer engine optimization, the practice of improving visibility in AI-generated answers): prompts that show what ChatGPT says when a buyer asks about your category.

What makes a ChatGPT SEO prompt reliable?

A reliable SEO prompt gives ChatGPT the raw material, names the output format and forbids guessing. Prompt engineering is the practice of structuring natural-language inputs to get a specified output from a generative model. For SEO, three habits do most of the work:

• Paste the data. ChatGPT cannot crawl your site or read your Search Console. Give it the keyword export, headings, robots.txt or draft.
• Ask for structure, not prose. Clusters, outlines, tables and flags are easy to check. Finished articles are not.
• Allow "I don't know". Add "If you lack the data, say so instead of estimating." A 2025 paper by Kalai and colleagues argues that models hallucinate because training and evaluation reward guessing over acknowledging uncertainty, so you have to ask for the admission.

Google's position is the same in spirit: generative AI output may contain inaccuracies, and Google calls it critical to fact-check and review AI-generated content before publishing.

Which prompts help with keyword research and clustering?

Use ChatGPT to group, label and expand keywords you already have; never to supply volumes. It is quick at sorting 300 phrases by intent and spotting near-duplicates. Take every number from Search Console or a keyword tool.

1. Cluster by intent

Here is a keyword list for [TOPIC]: [PASTE LIST].
Group it into clusters by search intent (informational, commercial,
transactional, navigational). For each cluster, name the one page
that should target the whole group. Do not give search volumes.

Check: search each cluster's head term. If Google shows different page types for two phrases, split them.

2. Expand a seed into buyer phrasings

Take "[SEED KEYWORD]" and write 15 phrasings a real buyer might type:
questions, "vs" comparisons, "for [use case]" and price wording.
Mark which sound commercial. Do not estimate volume.

Check: look each phrase up in Google Keyword Planner or Search Console before building a page.

3. Spot cannibalization

These page titles and URLs target similar topics: [PASTE LIST].
Which pairs likely compete for the same queries? For each pair,
say whether to differentiate or merge, and why.

Check: in Search Console, see whether two of your URLs earn impressions for the same query.

4. Read the intent behind a results page

Here are the titles and first headings of the top 5 results for
"[KEYWORD]": [PASTE]. What does the searcher want? What format
dominates (list, guide, tool, product page)? What is the gap?

Check: paste real results, not memory. ChatGPT's recollection of a results page is unreliable.

Which prompts help with briefs and outlines?

Ask for structure and gaps, not paragraphs. A brief prompt should name the keyword, the reader and the competing pages, then return headings and open questions. You still make the argument; ChatGPT just keeps you from forgetting a section.

5. Brief from one keyword

Build a content brief for "[KEYWORD]": reader, intent, 5-7 H2s phrased
as questions, one question competitors leave unanswered, and a
one-sentence angle that is not a generic explainer.

6. Competitor gap outline

Here is the text of the top 3 pages for "[KEYWORD]": [PASTE].
List what all three cover, what none covers that a buyer would still
ask, and an outline that fills the gap.

7. Answer-first openers

For each H2 below, write a 40-60 word opener that answers the
heading's question directly before any detail: [PASTE HEADINGS].
Use only facts I provide here: [PASTE FACTS].

Check: delete any number the opener contains that you did not supply.

8. FAQs from People Also Ask

Here are People Also Ask questions for "[KEYWORD]": [PASTE].
Pick the 5 most useful for a buyer researching [TOPIC] and draft
2-3 sentence answers, flagging any claim that needs a source.

9. Title and meta options

Write 8 title tags under 60 characters and 4 meta descriptions under
155 for "[TOPIC]" targeting "[KEYWORD]". Each must say what the reader
gets. No clickbait.

Check: Google says it may rewrite title links and truncates them to fit the device, so judge titles by what survives in the results.

Which prompts help with on-page and technical work?

These work when you paste the actual file and treat the answer as a first pass. ChatGPT cannot fetch your live site, so it reviews only what you give it. Run the real validators afterwards.

10. Internal link suggestions

Here is a paragraph from my page on [TOPIC]: [PASTE]. My other pages
cover [LIST TITLES AND URLS]. Where would a link help the reader?
Give the anchor text and the target URL from my list only.

11. FAQPage JSON-LD

Write FAQPage JSON-LD for these question and answer pairs, using
exactly the wording below: [PASTE Q&A]. Output only the code.

Check: validate in the Schema Markup Validator, or generate the markup with our AI Schema Generator. Every answer in the markup must appear on the page.

12. robots.txt review

Here is my robots.txt: [PASTE]. Which rules could block product,
blog or category pages, or an AI search crawler such as OAI-SearchBot?
Quote each rule you flag.

Check: Google notes robots.txt tells crawlers which URLs they can access and is not a way to keep a page out of Google, so confirm any "fix" in Search Console.

13. Heading structure audit

Here are all headings on a page in order: [PASTE H1-H6 LIST].
Flag skipped levels, duplicate H1s and headings that do not match
the section beneath them.

14. Redirect map review

Here are old URLs mapped to new ones: [PASTE]. Flag any where the
new page has a different intent from the old one, and any chain
of more than one hop.

Check: Google treats a permanent redirect as a signal that the target should be canonical, so test the real status codes with a crawler.

Which prompts help edit for quality?

Run editing prompts last, after the argument is settled. They cut, tighten and surface gaps; they should never be where your opinion comes from. "SEO text" generated in one shot is the fastest way to publish what Google calls scaled content abuse: many pages generated mainly to manipulate rankings, not help users. Generative AI tools are named as an example.

15. Answer-first rewrite

Rewrite this paragraph so its first two sentences answer the heading
above it, then add detail. Change no facts: [HEADING + PARAGRAPH]

16. De-fluff pass

Remove filler, hedging and repeated ideas from this text without
deleting any claim. List what you cut: [PASTE TEXT]

17. Claim-vs-source check

Here is a paragraph with a statistic or claim: [PASTE]. Does it need
a source? Describe the kind of source that would back it (study,
government data, vendor documentation). Do not invent one.

Check: find the source yourself. A fabricated citation is worse than none.

18. Interview me for first-hand detail

Read this draft: [PASTE]. Ask me 8 questions whose answers would
add first-hand experience, a real example or an opinion this page
lacks. Ask one at a time.

Google's guidance asks whether content shows first-hand expertise. Answer the questions honestly and the draft gets something no model has. Volume is not a plan: when we audited our own archive in September 2026, about half of 496 posts were unrelated to AI visibility and had no clicks in 90 days, so we retired 234 and merged 27.

How do you use prompts to check what ChatGPT says about your brand?

Ask unbranded buyer questions in a clean session, several times each, and log who gets named. A single answer is a sample of one. Start with prompt 19, because it is the closest to what a buyer does.

19. Category recommendation (the real test)

I need [CATEGORY, e.g. "an invoicing tool"] for [USE CASE].
What are the top options, and why would you pick each?

20. Constraint question

Which [CATEGORY] would you recommend for a [COMPANY TYPE] that needs
[REQUIREMENT]? Give a ranked shortlist with one reason each.

21. Brand perception (tests recall, not recommendation)

What do you know about [BRAND]? What does it do, who is it for, and
what would you tell someone deciding whether to use it?

22. Source check

What sources would you point to if someone wanted to verify what
you just said about [BRAND]?

23. Objection surfacing

What are the main downsides of [BRAND] compared with alternatives
in [CATEGORY]?

Prompts 21 to 23 name your brand, so they test accuracy: wrong prices, stale features, missing sources. Only 19 and 20 test whether you are recommended. That gap is common: in AEOeye's audit records (127 audits, June to September 2026), ChatGPT named the audited brand in 8 of 161 buyer-style answers (5.0%). It is a small, self-selected sample with repeat audits, not a market rate. More in our measurement hub.

How many runs do you need?

Three runs show whether an answer flickers; about seven give a usable rate. SparkToro's study (600 volunteers, 2,961 runs) found a below 1-in-100 chance that ChatGPT or Google's AI returns the same brand list twice. A 2026 preprint (Swiss-market prompts, four engines) found that 7 runs per prompt brings the 95% interval on a brand's detection rate to about ±16 points.

Run 19 and 20 at least three times each in a logged-out window or temporary chat, so your own history cannot flatter you. Log four things per run: named or not, linked or not, which competitors appeared, which domains were cited. In that preprint, ChatGPT searched the web for only some queries and returned zero citations in 57.8% of runs, so also note whether it searched.

Want a faster first look? AEOeye's free audit asks ChatGPT three unbranded buyer questions live and shows the full answers and competitors named. For hundreds of prompts on a schedule, use a dedicated tracker; our comparison lists them. For the manual route in full, see free AI visibility checkers; for what moves the answers, ChatGPT SEO.

How do you verify ChatGPT's SEO output?

Verify every output that contains a number, a name or a "fact", and match each prompt type to its source of truth. ChatGPT is fluent when wrong, so the error rarely looks like an error. Use this table.

Output	Typical mistake	Check it against
Keyword ideas	Phrases nobody searches	Keyword Planner or Search Console
Volume or difficulty	Invented figures	A keyword tool, never ChatGPT
Competitor analysis	Summaries of pages it did not read	The pages themselves
Schema, robots, redirects	Plausible but invalid code	Validator, crawler, real status codes
Statistics in drafts	Fabricated or outdated sources	The primary source, opened by you
Brand answers	One lucky or unlucky run	3 to 7 repeated runs

Search Console is the best truth source you own: its Performance report shows clicks, impressions, click-through rate and average position, grouped by query. Use it to confirm that a keyword, cluster or merged page does what ChatGPT said it would.

For the wider workflow of where AI helps and where it hurts, read AI for SEO. The honest summary: pick prompts where a wrong answer is cheap to catch.