Long-Tail Expansion

Variables

Replace each placeholder before you run the prompt.

• {{seed_keyword}}The keyword to expand from.

  industrial laser marker

• {{industry_context}}Vertical, target buyer, common materials / applications / specs.

  Industrial part marking. Buyers: manufacturing engineers, quality leads. Materials: stainless 316L, anodized aluminum, plastic. Applications: serial numbers, 2D matrix codes, traceability.

Generate 100 long-tail keyword variants from the seed below. Group by funnel stage so they can be routed to ad groups directly.

SEED KEYWORD:
{{seed_keyword}}

INDUSTRY CONTEXT:
{{industry_context}}

GENERATE:
- 30 TOFU variants (informational / problem-aware): "what is X", "how to X", "X vs Y", "best practices for X"
- 35 MOFU variants (solution-aware, comparing): "X for [application]", "X for [material]", "best X for [use case]", "compare X options"
- 35 BOFU variants (vendor-aware, ready to buy): include part-number patterns, dimensional specs, brand prefixes, "buy X", "X price", "X for sale [region]"

For each variant, include match type recommendation (exact / phrase). Default BOFU to exact, MOFU to phrase, TOFU to phrase or broad with tight negatives.

Do NOT generate consumer variants ("for home", "for kids", "DIY") — this is a B2B account.
Do NOT pad to 100 if the seed doesn't support it — return fewer high-quality variants instead of 100 weak ones.

Expected output shape

Up to 100 keyword variants in 3 funnel-stage buckets, each with a match-type recommendation.

Why we wrote it

Long-tail expansion is where Smart Bidding falls down for B2B — the model needs the right surface area, but most expanders surface consumer queries. This prompt encodes the right bias.

How to use

1. Open Claude or ChatGPT. The recommended model for this prompt is claude-sonnet-4-6 — opus when the prompt requires deep reasoning, sonnet for the rest.
2. Replace every {{variable}} with content specific to your account. The examples above are starting points, not templates to ship as-is.
3. Paste the prompt and run.
4. Read the output against the expected shape above. If the model produced a structurally different response, re-prompt rather than accept the drift.