How to Bucket AI Tracking Prompts by Commercial Intent and Buying Stage

Why bucket prompts at all?

An overall visibility score mixes very different questions. "What causes high employee turnover?" and "Best HR software for a 200-person company" are both relevant to an HR software company. But being cited for the first builds awareness, while being named for the second puts you on a shortlist. They respond to different work, too. Educational content drives the first, while reviews, comparisons and pricing pages drive the second. Buckets tell you where you are strong, where you are missing, and what to do about it.

What tagging scheme works?

The expected-answer tag is often skipped, but it matters. For an explanation prompt, success is being cited. For a list prompt, success is being named, ideally near the top. For a brand-facts prompt, success is accuracy.

What are examples of top-, mid-, and bottom-funnel prompts for AI search?

Using an illustrative expense management software company:

Top of funnel (problem and early solution)

• "Why do employee expense reports take so long to process?"
• "How do companies stop duplicate expense claims?"
• "What is a corporate card program and how does it work?"

Middle of funnel (solution and shortlist)

• "Best expense management software for a 200-person company"
• "Expense tools that integrate with NetSuite"
• "[Competitor] alternatives for mid-sized companies"

Bottom of funnel (decision)

• "[Your product] vs [competitor] for multi-entity companies"
• "[Your product] pricing per user"
• "Is [your product] SOC 2 compliant?"
• "How long does it take to implement [your product]?"

Post-purchase and renewal

• "How do I set up approval rules in [your product]?"
• "Is [your product] worth renewing, or should we switch?"

For a local service, such as a dental practice, the same structure applies. Top: "Why do my gums bleed when I floss?" Middle: "Best family dentist in [town]". Bottom: "Does [practice] take [insurer]?" and "[Practice] Saturday appointments".

How do you assign stage and intent quickly?

Signal words make most prompts easy to tag:

A spreadsheet formula or a simple script can suggest tags from these signals, and a person can review the suggestions in a few minutes. Review matters, because some prompts carry signals from two stages. "Best way to cut expense processing time" reads like a shortlist question but is often asked by someone still defining the problem.

How to track AI citations by query intent instead of overall visibility

Record, for every run, which domains are cited, then summarise by bucket:

• Citation share by bucket: your share of citations in informational, commercial and transactional prompts separately.
• Source types by bucket: early-stage answers often cite publishers, reference sites and guides. Shortlist answers lean on review platforms and "best of" lists. Decision answers draw on vendor pricing and documentation pages, and review profiles.
• Competitor citations by bucket: which competitors' pages are cited where you are not.

This turns citation data into actions. If you are cited for informational prompts but absent from commercial ones, the gap is usually in comparison content, review platforms and the lists that shortlist answers rely on, not in more educational articles. If decision prompts cite third-party pages about your pricing, your own pricing page may be unclear or hidden.

How many prompts does each bucket need?

With three runs per prompt, 10–15 prompts per bucket per engine, 30–45 answers, is a minimum. At that size a bucket's mention rate is only good to about ±15–18 points, so treat month-to-month moves under about 25 points as noise, or pool two or three months or several engines before comparing buckets. Smaller buckets swing with a handful of answers. If you have many personas or product lines, keep the stage and intent buckets well filled first. Add finer cuts only where there are enough prompts.

What pitfalls should you avoid?

• Branded prompts in the main score. Prompts that name your brand almost always mention you. Keep them in their own bucket, and use them to measure accuracy.
• Prompts written by marketing only. Draw from sales calls, support tickets and search data, so buckets reflect real buyer language.
• Unbalanced buckets. Fifty informational prompts and five commercial ones produce a score dominated by education.
• Changing prompts without notes. Record changes by bucket so trends stay comparable.
• Treating stage as fixed. Some prompts sit between stages. Tag the most likely stage and move on, rather than debating edge cases.

Choosing prompts in the first place is covered in how to choose the right prompts to track for AI visibility. Whether a published prompt set has been cherry-picked is covered in how to detect cherry-picked prompts in an AI search case study.

What does a bucketed report look like?

A one-page summary can show, for each stage:

• mention rate on unbranded prompts, by engine;
• your citation share and the top cited domains;
• the main competitor named;
• accuracy issues found, for branded and decision prompts;
• one recommended action.

Consider an illustrative result. The expense software company appears in 40% of top-of-funnel answers, mostly as a cited guide, and 12% of shortlist answers. It appears in 70% of decision answers, where its pricing is described correctly in only half. The actions follow directly: work on review platforms and trade lists for shortlist prompts, and fix the pricing page for decision prompts.

What do people ask about bucketing prompts?

Should buckets follow our marketing funnel or the buyer's journey?

The buyer's journey. Your funnel stages can map onto it, but prompts should reflect how buyers think.

How should bucket results be weighted?

By commercial value. Shortlist and decision buckets usually deserve more weight in a headline figure, but report each separately too.

Can one prompt belong to two buckets?

Give it one primary stage and intent. Double-counting makes trends harder to read.

How often should buckets be reviewed?

Quarterly, when you review the prompt set as a whole.