AI Keyword Tool: Cluster Keywords by Intent and Funnel Stage

What does this AI keyword tool do?

It does three jobs in one run: it classifies every keyword by search intent and funnel stage, it clusters the list into topics, and it turns each cluster into a hub page with spoke pages and sections. The reference for the format is the Jev tool on AnswerSocrates, which does the same three jobs behind a sign-in and a credit meter using a decision model. This one uses rules, runs free, and keeps your keyword list on your machine. The trade is judgement for transparency: you can read exactly why a keyword got its label, and the ambiguous ones are flagged rather than guessed.

How does it classify search intent and funnel stage?

By the words in the keyword, in a fixed order: buying words first, then navigation words, then comparison words, then question words, with informational as the default. Each match adds to the confidence score, and a keyword that carries both a comparison word and a question word is marked to check by hand.

Signal in the keyword	Intent	Funnel stage	Examples of the trigger words
Buying or booking word	Transactional	BOFU	buy, price, cost, hire, quote, near me, free trial, order
Navigation word, or a brand term on a short keyword	Navigational	BOFU if branded, else MOFU	login, sign in, official, website, contact, your brand names
Comparison or vendor word	Commercial	MOFU	best, top, review, vs, alternative, tool, software, agency, consultant
Question or learning word	Informational	TOFU	what, how, why, guide, meaning, examples, tips, checklist
No signal at all	Informational (inferred, 0.6)	TOFU	a plain noun phrase such as standing desk

Two flags ride alongside: local when the keyword carries near me or a city name, and branded when it contains one of the brand terms you typed in. Rows longer than twelve words, or containing prompt language, are flagged as noise, because Search Console increasingly records pasted AI prompts as queries, and a prompt is not a page.

How does keyword clustering turn into a hub and spoke plan?

The clusterer first drops the theme word that most of your list shares, then groups keywords that share two or more significant words or overlap by a third or more, and finally picks the hub of each cluster as the informational keyword with the most impressions. Dropping the theme word matters more than it sounds: on this site the word SEO appears in most queries, and without that step the first run put 1,406 keywords into one cluster called seo search. With it, the same list produced 247 real clusters.

Inside a cluster, a keyword becomes a spoke page when it sits at a different funnel stage from the hub, or when it carries a qualifier that appears three or more times in the cluster, which is the sign that Google treats it as its own topic. Everything else becomes a section of the hub page. That is the same hub and spoke logic as topical authority done by hand, with the arithmetic done for you. Spokes are capped at eight per cluster so the plan stays buildable.

What did it find when we ran it on this site?

Of the 2,197 unique queries that showed lawrencehitches.com in Google over the 28 days to 25 September 2026, 74% were top of funnel, 25% middle and 2% bottom. That is the honest shape of an informational site: the demand is people learning, and the buying queries are a rounding error. It also explains why the money pages on this site are an authority problem rather than a content one.

Cluster	Keywords	Impressions	TOFU / MOFU / BOFU	Hub the tool picked
google bing	93	631	50 / 43 / 0	which is better google or bing
tracking ai	90	2,914	58 / 32 / 0	ai mode tracking
source utm	62	2,152	56 / 6 / 0	utm source chatgpt
enterprise best	57	2,400	0 / 54 / 3	enterprise seo software
data structured	53	338	47 / 6 / 0	structured data for ai search
claude skill	50	1,013	39 / 11 / 0	claude seo skills
claude referral	35	540	33 / 2 / 0	claude referral
enterprise platform	29	869	0 / 28 / 1	enterprise seo platform

Two things the run taught us before it shipped. The first version snowballed everything into one cluster, which is why the theme-word step exists. And 36 of the queries were pasted prompts rather than searches, which is why the noise flag exists. Real output beats an invented example, and it also audits your own estate on the way through.

How do you plan TOFU, MOFU and BOFU content from the output?

One hub page per cluster, spoke pages only where the tool found a distinct stage or a repeated qualifier, and sections for the rest. Build the middle of the funnel first if the site already ranks for the top: comparison and vendor queries are where a reader is choosing, and a hub that only teaches sends them elsewhere to decide. If the list is mostly bottom of funnel, the missing layer is usually the explainer that earns the links. The keyword research workflow covers how to source the list, and chunky middle keywords covers the band where most of these clusters sit.

What are the limits?

It does not know search volume unless you bring it, it does not read the SERP, and it does not understand a keyword the way a person does. Autosuggest reflects what Google offers people typing, not how many type it. Rules mis-read sarcasm, product names that look like verbs, and languages other than English. Treat the export as a first pass that removes the tedium, not as a plan you ship unread. For the model-assisted version of the same job, the Claude topical map workflow and the query-templates plugin on the Claude SEO plugins page take the CSV from here.

Frequently asked questions

Is this AI keyword tool free?

Yes. It is free, there is no account and no credit system. Paste mode runs entirely in your browser and sends nothing anywhere. Seed-topic mode makes one request per suggestion query to this site, which relays it to Google's public autosuggest endpoint and returns the strings.

Where do the keywords come from in seed-topic mode?

From Google autosuggest: the seed, the seed with each letter of the alphabet, and the seed with question and buying prefixes. That is what Google offers people typing your topic, not a volume database. Treat the list as demand shape, and bring volumes from your own source if you need them.

How is this different from an LLM keyword tool?

It uses fixed rules, so the same keyword gets the same label every time and you can read why in the confidence column. An LLM tool can read ambiguous keywords better and can also invent a label. If you want both, run this first, export the CSV, and hand the low-confidence rows to Claude.

Can I upload a Search Console export?

Yes. Export Queries from Search Console as CSV and upload it. The tool reads the query, clicks, impressions and position columns, so clusters are ranked by real impressions and the hub of each cluster is the keyword Google already shows you for.

What does the confidence score mean?

It is rule confidence, not probability of being right. 0.7 to 0.95 means one or more explicit signals matched (a buying word, a comparison word, a question word). 0.6 means no signal matched and the keyword was read as informational by default. 0.55 means signals pointed two ways, and those rows are flagged to check by hand.