Classify an account into one of your defined segments
Assigns every row in your table to exactly one of your named segments, with confidence and a short evidence trail. The exact-string rule means outputs feed lookups, filters, and campaign routing without normalization, and UNCLASSIFIED gives ambiguous rows a safe home instead of contaminating a segment.
You are a strict single-label classifier running inside an enrichment pipeline. Your only job is to assign one account to exactly one segment from a defined list. You are not creative. You do not blend segments, invent new ones, or hedge with 'could be either'. Ambiguity has a designated output, and it is not a guess.
You will be given two inputs: (1) a list of segment definitions, each with a name and 1–3 sentences of criteria, and (2) the account's data — typically website text, an industry field, employee count, and any enrichment notes provided.
Classification rules:
- Evaluate the account against every segment's criteria before answering, in the order the segments are listed.
- Match on evidence in the provided data only. General knowledge about the company name is off-limits — if the data doesn't show it, it doesn't count.
- If the account plausibly fits two segments, choose the one whose criteria it matches more completely; if genuinely equal, choose the one listed first and lower your confidence.
- If the account matches no segment, or the data is too thin to evaluate the criteria, classify it as UNCLASSIFIED.
OUTPUT CONTRACT — follow exactly:
Return a single JSON object:
{"segment": "<exactly one segment name from the provided list, character-for-character, or UNCLASSIFIED>", "confidence": "<high | medium | low>", "evidence": "<under 20 words: the specific data points that drove the decision>"}
- Return ONLY the JSON. No markdown fences, no commentary.
- The segment value must be copied exactly from the provided list — never abbreviated, never reworded, never a new label.
- UNCLASSIFIED with low confidence is a correct and welcome answer for thin data. A confident wrong label is the only failure.How to use it
- 1
Copy the prompt into Claude, ChatGPT, or any LLM — or into a Clay AI column.
- 2
Write your segment definitions once (name plus 1–3 sentences of criteria each) and paste them into the prompt or reference them from a table-level column; then map your account-data columns as the second input.
- 3
Run 30 rows you can classify by eye and compare — disagreements usually mean a criteria sentence is vague, so fix the definition, not the prompt.
- 4
Route each segment value to its own campaign, and send UNCLASSIFIED plus low-confidence rows to manual review or a generic track.
Best practices
Keep segments to 3–7; past that, adjacent definitions blur and confidence drops across the whole table.
Write criteria as observable facts ('sells to restaurants', 'over 200 employees'), not aspirations ('growth-minded companies') — the classifier can only match what data can show.
Audit the evidence field on a sample weekly; it's the fastest way to catch a definition drifting from what you meant.
Example: what this looks like in practice
A RevOps manager runs three campaign tracks: 'agency-partners', 'in-house-teams', and 'marketplaces'. She writes three-sentence criteria for each, pastes them into this prompt in a Clay AI column, and maps website text plus employee count as account data. Across 8,000 rows: 5,900 classify high-confidence, 1,300 medium, 800 UNCLASSIFIED. Spot-checking evidence strings, she finds 'white-label' language pulling marketplaces into agency-partners, tightens one criteria sentence, and re-runs the mediums. Each segment then flows to its own Smartlead campaign automatically — a routing job that previously took a VA two days per list and produced more errors.
Best fit
This prompt is one gear in a bigger machine. We orchestrate 20+ tools into outbound systems our clients own — and guarantee the results.
Apply for a Pilot Spot → →Frequently asked questions
Yes, when you treat it as strict classification rather than open judgment: named segments, written criteria, evidence-only matching, an exact-string output rule, and an UNCLASSIFIED escape hatch. Under those constraints agreement with human classification typically lands above 90% on well-defined segments — and the confidence field tells you exactly where to look for the rest.
More personalization at scale (clay & claygents) prompts
Generate a cold email first line from scraped research data
The workhorse of AI-personalized outbound: converts whatever research your enrichment waterfall produced into one 12–22-word observation-led first line per row. The swap-test rule and fact-invention ban are what separate lines that read researched from the AI slop prospects now delete on sight.
Turn a Claygent website scrape into a personalized one-liner
Produces a sub-25-word, website-specific opening line for every row in a Clay table — no human in the loop. The output contract forces plain text only and a NOT_FOUND fallback, so unreachable sites and thin homepages return a filterable sentinel instead of a hallucinated compliment that ends up in a prospect's inbox.
Analyze a pricing page for sales-relevant intelligence
Extracts a structured pricing profile from any pricing page: model, tier count, exact displayed prices, free-offer type, and a one-line read on their sales motion. At scale it segments prospect lists by how companies charge — which predicts budget, buying process, and which of your angles will land.