Best AI Prompts for Keyword Research (2026)
Prompts for keyword research#
Five starting points#
Act as an SEO analyst. Below is my keyword list for [DESCRIBE YOUR SITE AND AUDIENCE], with monthly search volume from [PASTE YOUR TOOL NAME] where I have it. Group the keywords into clusters that one page could answer, and return a table with one row per cluster: cluster name, the keywords in it, the dominant intent (informational, commercial investigation, transactional or navigational), and a one-line reason. Use only the volumes I supplied; do not estimate any number. Put keywords I did not give volumes for in a separate list. <keywords> [PASTE YOUR KEYWORD EXPORT] </keywords>Act as a content strategist. Here are my existing pages with their target topics [PASTE URL AND TITLE LIST] and my keyword clusters [PASTE CLUSTERS]. Map each cluster to the one existing page that should own it, or mark it "new page needed" with a working title. Flag clusters that two or more pages could claim, and say which page should win and why. Do not add keywords I did not provide.Act as a search researcher. Starting from these seed topics [PASTE YOUR SEED TOPICS] for [DESCRIBE YOUR READER], list 40 candidate phrases a reader might type, grouped by the question behind them. Label every phrase "unverified candidate": I will check each one in Keyword Planner and Search Console myself. Do not state volumes, difficulty or trends.Act as an SEO auditor. This is a Search Console Performance export of queries, pages, clicks, impressions and average position for the last [STATE DATE RANGE] [PASTE YOUR EXPORT]. Find queries where two or more of my URLs appear, show the URLs side by side with their clicks and impressions, and recommend for each case whether to merge, differentiate or internally link the pages. Quote the rows you based each call on.Act as a search intent specialist. For each query in [PASTE YOUR QUERY LIST], I have pasted a short note on what the top results look like: page types and angles [PASTE YOUR SERP NOTES]. Classify the intent from those notes only, say which page format would fit (guide, comparison, product page, tool), and say "notes insufficient" where the notes do not settle it.What to add to any keyword research prompt#
- The data itself, pasted in. Google says Keyword Planner gives estimates of monthly searches and the Search Console Performance report shows your queries, impressions, clicks and average position. A chat model has none of this unless you supply it.
- Long lists first, instructions last. Anthropic advises placing long documents and inputs near the top of the prompt, above the query and instructions, and wrapping content in XML tags such as
<keywords>so it is not confused with instructions. OpenAI also recommends XML tags to mark where reference content starts and stops. - The exact output shape: a table with named columns, one row per keyword or cluster. OpenAI and Google both advise stating the format explicitly.
- Your intent labels, defined in a line each, plus two or three example rows. Google lists examples as a way to steer the model, and Anthropic recommends 3 to 5 for best results.
- What to do with gaps: a rule such as "put keywords without a volume in a separate list". Google notes that Keyword Planner may not show data for keywords with very low volumes.
- Your reader. Google's helpful-content guidance asks whether a reader will leave feeling they have learned enough to achieve their goal, so say who is searching and what they are trying to do.
A worked example#
Prompt 1 filled in with realistic inputs:
Act as an SEO analyst. Below is my keyword list for a small online shop selling
loose-leaf tea, aimed at UK home drinkers, with monthly search volume from
Google Keyword Planner where I have it. Group the keywords into clusters that
one page could answer, and return a table with one row per cluster: cluster
name, the keywords in it, the dominant intent (informational, commercial
investigation, transactional or navigational), and a one-line reason. Use only
the volumes I supplied; do not estimate any number. Put keywords I did not give
volumes for in a separate list.
<keywords>
how to brew oolong tea | 590
oolong tea benefits | 1,300
buy oolong tea online | 320
best oolong tea | 480
oolong vs green tea |
loose leaf tea subscription | 210
</keywords>
A good result for this task:
- Every keyword appears once, either in a cluster or in the "no volume" list, with none added or dropped.
- Volumes in the output match your input exactly; any figure not in your paste is a failure.
- Each cluster has a stated intent with a reason you can check against the live results.
- Clusters that overlap (for example "best oolong tea" and "buy oolong tea online") are separated by intent or flagged for a decision.
- Cluster names are short enough to become page titles.
The full prompt library is on our prompt library page.
Limits of this guide#
The vendor guides explain how to supply data and structure a prompt. None of them tests keyword research, and we did not test whether AI clusters match the ones a specialist would build. Google's pages describe what Keyword Planner and Search Console report; they do not say how accurate those figures are, and we could not confirm a data source for Keyword Planner volumes. The intent labels in our prompts are our own working categories, not a Google taxonomy.
Sources & Further Reading#
All sources fetched on 2026-10-22. The OpenAI page was read at its original URL; the Internet Archive copy was not reachable.
| Claim | Source | Section or date |
|---|---|---|
| Place long inputs above the query; wrap content in XML tags; 3 to 5 examples work best | Prompting best practices (Anthropic) | Long context, XML tags and examples sections; undated |
| Add data the model lacks; XML tags mark where reference content starts and stops; state the output format; use varied examples | Prompt engineering (OpenAI) | Context, delimiters, output and few-shot sections; undated |
| Include context in the prompt; state output format; use few-shot examples; set constraints | Prompt design strategies (Google Gemini API) | Undated |
| Keyword Planner gives estimates of monthly searches; very low-volume or sensitive keywords may not be forecastable | About Keyword Planner (Google Ads Help) | Undated |
| Performance report shows queries, clicks, impressions, average position; results vary by time, place, device and history | Performance report (Search Console Help) | Undated |
| Reader should leave having learned enough; content made mainly for search visits or in volume across many topics is a warning sign | Creating helpful, reliable, people-first content (Google Search Central) | Undated |
Need a different task?
Browse all 36 tasks and 180 prompts, with search and level filters.
Frequently asked questions
What are the best AI prompts for keyword research?
Use AI to sort and map keywords, not to measure them. Paste your own keyword export or Search Console queries, ask for clusters by intent, and treat any volume or difficulty the model states as invented unless it came from your data.
Can ChatGPT or Claude tell me search volume?
Not by itself. A chat model has no live search-volume feed. Google Keyword Planner and the Search Console Performance report are the Google sources for search estimates and your own impressions. Export numbers from them and paste them into the prompt.
How do I get AI to cluster keywords by intent?
Paste the full keyword list in a delimited block, name the intent labels you want, and ask for one row per keyword with its cluster and a one-line reason. Then open the results page for a sample of each cluster, because the model has not seen the live results.
By TheBestSuperintelligence Editorial Team. Last reviewed .
How we sourced this. Model names, features and limits change often. The prompts below are our own untested starting points, and we did not measure which wording performs best. Facts are as printed on each source page on the date shown. Methodology
