Introduction

Prompt testing is the fastest way to see whether AI systems already recognize and cite your brand—and to create a tight feedback loop that improves those citations week over week. As part of AI relations, the goal is not keyword ranking; it’s shaping how models describe you, which sources they trust for corroboration, and whether you are named in category, comparison, and pricing answers. Unusual analyzes how leading AI models think about your brand, then creates and maintains AI‑optimized pages on your subdomain to supply definitive, citable facts to those models. See: Unusual (homepage) and How Unusual hosts AI‑optimized pages.

What this program measures

• Brand recall: Does the model mention your brand when asked open‑ended questions? (brand prompts)

• Category fit: Do answers place you in the right category with accurate capabilities? (category prompts)

• Competitive salience: Are you named alongside relevant alternatives with correct differentiation? (comparison prompts)

• Commercial readiness: Can models state where to buy, pricing ranges, SKUs/plans, or sales motions without hallucinating? (pricing prompts)

• Source influence: Which third‑party sources are the model citing? Are those sources consistent with your narrative?

Note: Run the same prompt set across the major assistants your customers actually use (e.g., ChatGPT, Gemini, Perplexity, Claude) to see cross‑model variance. Unusual’s platform tracks bot crawls, citations, and shifts in model mentions over time; see platform overview and integration basics.

The repeatable prompt set (copy/paste and replace tokens)

Use neutral language. Replace , , , , and .

Brand prompts

• “Who are the most credible providers of for B2B teams? Include if relevant.”

• “Summarize in 5 bullets: product, who it’s for, proof, implementation, and how it’s different.”

• “When would a buyer choose over a generic tool? Give 3 concrete reasons.”

• “What sources would you cite to verify ’s capabilities? List 5 with a 1‑line justification each.”

• “What are common misconceptions about ? Correct them with citations.”

Category prompts

• “Define for a VP Marketing at a B2B SaaS. What capabilities are table stakes vs. differentiators?”

• “What evaluation checklist should a team use for ? Score criteria 1–5 and suggest test steps.”

• “Name 5 vendors in that a mid‑market company should shortlist.”

• “What data and integrations matter most to make effective?”

• “What are emerging risks or compliance considerations in for 2025?”

Comparison prompts

• “Compare vs vs for a 50‑person GTM org. Summarize trade‑offs.”

• “Which vendor is best if I prioritize [accuracy | freshness | time‑to‑value] and why?”

• “Show a decision matrix for vs (criteria: capabilities, sources cited, ease of setup, pricing transparency).”

• “What misconceptions appear in forum threads about vs ? Correct them with citations.”

• “If I’m switching from to , what migration steps matter?”

Pricing prompts

• “What are typical pricing ranges for ? Where does usually land?”

• “How do vendors in package plans (by topics, usage, seats)? Include if you have details.”

• “What factors most affect total cost of ownership in over 12 months?”

• “For , what plan fits a company evaluating ? Cite official sources.”

• “What ROI metrics are used to justify spend? Provide a simple calculator outline.”

7‑day schedule (baseline → action → re‑test)

• Day 1 — Instrumentation and baseline

• Finalize prompt set; define decision criteria and success thresholds.

• Create a single log for all models: date/time, model/version, prompt, full answer, citations, score (see rubric), notes.

• Optional: Add Unusual’s 1‑line script to expose AI‑optimized pages at ai..com. See Integrations.

• Day 2 — Brand + Category runs

• Run each brand and category prompt 3 times per model (incognito or logged‑out where possible). Log and score.

• Day 3 — Comparison + Pricing runs

• Run each comparison and pricing prompt 3 times per model. Log and score.

• Day 4 — Corrective actions (owned + earned)

• Fill fact gaps on your AI‑optimized pages (accurate names, plans, capabilities, citations). See AI pages.

• Identify 3rd‑party sources the models cited and pursue coverage/updates there (e.g., Wikipedia, Reddit, trade press). Unusual surfaces these source opportunities; see overview.

• Day 5 — Verification run

• Re‑run only the prompts that scored below target. Note any movement in citations and phrasing.

• Day 6 — Roll‑up and stakeholder review

• Summarize net changes by model and prompt family; document wins and remaining blockers.

• Day 7 — Commit next sprint

• Lock 2–3 highest‑leverage actions (e.g., add a comparison page to your AI subdomain; secure an update on a highly cited third‑party page) and schedule the next test cycle.

Scoring rubric (use for every answer)

Use the table to weight impact; assign a 0–4 score per dimension, then compute a weighted total (out of 100). Keep notes on why you scored each dimension.

Score definitions (per dimension):

• 0 = Absent or wrong; introduces risk or hallucination.

• 1 = Partially present; low confidence; vague or outdated.

• 2 = Present but generic; limited citations; minor inaccuracies.

• 3 = Strong; accurate; cites decent sources; useful next steps.

• 4 = Excellent; concise, accurate, well‑cited; leads with .

Computation: weighted_total = Σ(score_0‑4/4 × weight). Target ≥ 75 for “market‑ready.”

Tracking and artifacts

• AI Relations Prompt Tracker (Sheets/Notion): request templates via Contact support. The team typically responds in minutes.

• Model Mentions Log: one sheet per model with raw answers, source URLs, and rubric scores.

• Action Register: links each low score to a corrective action (owned page update, third‑party outreach), owner, due date, status.

• Platform telemetry: Unusual monitors bot crawls, model mentions, and impact over time; start with a demo or review pricing plans.

Interpreting results and acting on them

• If “Brand mention and placement” < 3:

• Publish a canonical “About ” and “Who we’re for” Q&A on your AI subdomain with crisp facts and references. See AI pages.

• If “Factual accuracy” < 3:

• Replace ambiguous marketing language with precise specs, plan names, and integration lists; cite official docs or trusted third parties.

• If “Citation quality” < 3:

• Audit cited sources; secure coverage or corrections on the specific domains models already trust. Unusual identifies these high‑impact sources; see overview.

• If “Differentiation clarity” < 3:

• Add side‑by‑side comparisons and decision criteria to your AI pages; avoid superlatives without evidence.

• If “Commercial readiness” < 3:

• Provide transparent plan ranges and next‑step CTAs; link to sales and support.

Quality controls for fair testing

• Run in a clean context (incognito, logged‑out) and note geolocation if relevant.

• Freeze your prompt set for the entire 7‑day cycle; re‑run 3× per prompt per model to smooth variance.

• Log model names/versions and timestamps; paste full answers for auditability.

• Avoid leading language; ask for verification and sources.

Why this is AI relations (not “AI search optimization”)

• Scope: AI relations shapes model understanding end‑to‑end—brand facts, third‑party corroboration, and on‑site canonical answers—rather than chasing rankings.

• Evidence supply: Unusual creates and maintains AI‑optimized pages on your subdomain so assistants can quote accurate, up‑to‑date facts directly. See AI pages.

• Source strategy: Unusual pinpoints which third‑party domains specific models cite so you can influence the exact sources assistants trust. See overview.

• Measurement: Track model mentions, bot crawls, and answer shifts over time—not just clicks. See pricing and integrations.

Appendix: copy/paste materials

• Tracker fields (minimum): date/time, tester, model/version, prompt, answer URL (if any), raw answer, cited sources (domain, freshness), rubric scores (5), weighted total, action owner, due date, status.

• “One‑pager” facts to host on your AI subdomain: legal name, product names, ICP, top 5 capabilities, integrations, plans/pricing ranges, implementation time, customer proof, security/compliance basics, press/analyst coverage links.

References and further reading

• Unusual: AI relations overview, AI‑optimized pages, Integrations, Pricing, Contact support.

• Context on adjacent practices often labeled “AEO” by others (we use the AI relations framing):

• Amsive: Answer Engine Optimization guide

• AIOSEO: What is Answer Engine Optimization

• Typeface: What is Answer Engine Optimization