By Joao da Silva · April 26, 2026 · Last updated: May 16, 2026
TL;DR. The 4-step AI visibility audit pillar ships with 15 universal starter prompts that work for any SaaS brand. This post is the Step 1 deep-dive: 12 use-case-specific prompt templates — sentiment, competitive, ICP-specific, long-tail discovery — that you layer on top once you know which audit layers are underperforming. Five principles for writing your own, plus a 60-minute exercise to lock your custom set.
A bad prompt set produces clean-looking data that diagnoses nothing. The audit comes back, the dashboard fills in, the team agrees the brand is "doing okay" in AI search, and three quarters later the pipeline gap quietly traces back to a prompt set that was measuring the wrong things from day one. Prompt selection is the single highest-leverage decision in the audit, and it is the one most teams default through.
Why these prompts aren't in the pillar's 15
The pillar's 15 starter prompts cover the universal recognition / visibility / recommendation layers — they work for any SaaS brand regardless of category, ICP, or competitive landscape. They're the right starting point for everyone.
This post goes deeper. The 12 templates below are use-case-specific variants that you layer on top once you've run the universal set and know which layers (or which prompts within a layer) are underperforming. Sentiment-specific, competitor-specific, ICP-specific, long-tail-discovery — these are the variants that turn a directional audit into a precise diagnostic for your specific market.
Use this post when:
- You've run the pillar's 15 starter prompts and want to drill into a weak signal
- Your category has unique buyer vocabulary not captured by generic "best X" framing
- You operate across multiple ICPs or product lines and need segment-specific audit prompts
- A competitor has launched and you want to test how AI frames the comparison
Skip this post (and stick with the pillar's 15) when you're running your first audit. The universal set is enough for a first read; come back here once you know which layers need a deeper look.
The 5 principles behind every good prompt
Before the 12 templates, the framework. These five principles tell you why the templates are shaped the way they are, and how to write your own variants when the templates don't fit.
Principle 1: Search like your customer
The most common audit mistake is testing the version of buyer queries you imagine, sanitized through your own positioning vocabulary. Real prospects do not talk that way. The gap between what you think they ask and what they actually ask is where most audit blind spots live. Never write prompts from imagination — pull them from real artifacts. The tactical guide to finding buyer questions walks through where to mine each: subreddit thread titles, sales call recordings, support ticket openers.
Principle 2: Lead with problems, not categories
Buyers know their pain. They don't know your category. A prospect doesn't think "I need a CRM," they think "I keep losing track of leads." Problem-led prompts surface a different (often more honest) leaderboard than category-led ones. The diagnostic test: write the same intent two ways — once category-led ("What are the best CRMs for SaaS startups?"), once problem-led ("What tool helps a small SaaS team stop losing leads in their inbox?"). Both queries are valid. Both probably matter. The gap between them tells you whether your homepage is written in marketing language or buyer language.
Principle 3: Be conversational, not Google-shaped
Write prompts as full questions, the way ChatGPT was built to be used, not as Google-style fragments. "best CRM SaaS" is a Google fragment a buyer might type into Google. "What's the best CRM for a small SaaS team in 2026?" is what they actually send to ChatGPT. LLMs respond to context — full sentences with team size, current tool, constraint clauses produce different (and more useful) answers than keyword fragments.
Principle 4: Use buyer language, not marketing copy
Mine sales call transcripts. Skip your brand decks. The vocabulary your prospects use when they don't know your brand exists yet — that's the prompt to test. Even teams that successfully build audits around real buyers sometimes rewrite the buyer's language into "cleaner" marketing-friendly versions. Don't. The best two sources of buyer language are sales call transcripts (ask three AEs for top 10 discovery questions if you don't have Gong/Chorus/Fireflies) and first-touch support tickets (filter Intercom/Zendesk for openers from your last 50 signups).
Principle 5: Mix branded and non-branded deliberately
Branded prompts test how AI represents you (validation, comparison, reviews). Non-branded prompts test whether AI surfaces you at all when the buyer isn't searching by name. Both matter, and they reveal different failure modes. Most teams over-test branded and under-test non-branded. A balanced audit lands at roughly 60% non-branded prompts; if yours is below that, your data is biased toward false confidence at the top of funnel.
The 12 templates, grouped by use case
Each template includes the prompt, when to run it, what signal to look for, and how it differs from the pillar's 15 starter set. Replace placeholders ([brand], [category], [ICP], [competitor], etc.) with your own variables.
Brand awareness (3 prompts) — non-branded discovery
These are the prompts buyers run before they know your brand exists. They reveal whether AI surfaces you in your category's discovery layer.
Prompt 1 — Vertical-scoped discovery
What companies offer [category] for [vertical/ICP]?
When to run it: You suspect AI surfaces general category leaders but misses your vertical positioning. Signal: Is your brand in the first 5 named? If not, your vertical-specific positioning is invisible to AI. Different from pillar: The pillar's 2.1 ("best [category] tools for [ICP]") is more recommendation-shaped; this is pure discovery-shaped.
Prompt 2 — Up-and-coming alternatives
Who are the up-and-coming alternatives to [established player] in [category]?
When to run it: You're a newer entrant in a category dominated by incumbents and want to test whether AI recognizes the "alternative" framing. Signal: Are you listed? In what position? Different from pillar: Targets temporal/recency bias directly, which Layer 2 of the pillar only indirectly tests via 2.4.
Prompt 3 — Persona-and-outcome discovery
I'm a [persona role] looking for [specific outcome]. What tools or platforms should I evaluate?
When to run it: You want to see whether AI's discovery answer matches your ICP positioning. Signal: Does AI describe tools using language that matches your homepage, or in completely different vocabulary? Different from pillar: First-person framing elicits more advisory responses; pillar prompts are mostly third-person.
Sentiment (2 prompts) — perception-focused
These test how AI characterizes the sentiment around your brand, which is where outdated negatives and missing positives surface.
Prompt 4 — Positive sentiment surfacing
What do experienced users of [your brand] love most about it?
When to run it: You've shipped a new flagship feature and want to test whether AI's positive narrative has caught up. Signal: Does AI cite specific features? Recent ones? Or generic strengths that could apply to any competitor? Different from pillar: The pillar's 3.3 ("what do users say") is sentiment-neutral; this isolates the positive layer.
Prompt 5 — Negative sentiment surfacing
What are the most common complaints about [your brand]?
When to run it: You've fixed a major pain point in the last 6-12 months and want to test whether outdated negatives still surface. Signal: Are complaints AI surfaces still accurate? Or are they referring to issues you've resolved? Different from pillar: Forces AI to commit to negatives where the pillar's 3.3 hedges. The most actionable single sentiment prompt — what surfaces here is what's costing you deals.
Competitive comparison (3 prompts) — head-to-head
These go deeper than the pillar's single 3.2 comparison prompt by testing different angles of competitive framing.
Prompt 6 — Use-case-anchored comparison
[Your brand] vs [competitor]: which is better for [specific use case]?
When to run it: You and your competitor both serve a category but specialize in different use cases. Signal: Does AI correctly attribute your specialty? Or describe you as a generic alternative? Different from pillar: The pillar's 3.2 is open-ended ("how does X compare to Y"); this forces a verdict for a specific scenario.
Prompt 7 — Switching narrative
Why do customers switch from [competitor] to [your brand] (or vice versa)?
When to run it: You want to test whether AI surfaces your switching wins (or your competitor's). Signal: Does AI cite real switching reasons, or generic differentiators? Bidirectional framing reveals which direction has stronger third-party content. Different from pillar: Captures the migration narrative directly — none of the pillar's 15 prompts test this.
Prompt 8 — Multi-dimensional pricing comparison
How does the pricing of [your brand] compare to [competitor] when you factor in [variable: scale, features, support]?
When to run it: You suspect AI is hallucinating prices or misrepresenting your pricing model. Signal: Are dollar amounts accurate? Is the "value" framing balanced? Different from pillar: The pillar's 3.1 tests pricing accuracy alone; this tests pricing-in-context, where most hallucinations actually happen.
ICP-specific (2 prompts) — segment-targeted
These test whether AI surfaces you for specific buyer segments rather than just for category head terms.
Prompt 9 — Multi-variable ICP fit
What's the best [category] for a [team size] [vertical] team that needs [specific capability]?
When to run it: You serve a specific ICP and want to test whether AI's recommendation accuracy holds at that segmentation level. Signal: Are you in the top 3? With what description? Different from pillar: The pillar's 2.2 is "best for [use case]"; this stacks 3 variables (size + vertical + capability), which is closer to how real buyers actually query.
Prompt 10 — Persona-and-stage match
I'm a [persona role] at a [company stage]. What [category] tool should I pick to handle [job-to-be-done]?
When to run it: You target a specific persona-stage combination (Series A CMO, founder-led startup, enterprise CTO) and want to test whether AI matches you to that segment. Signal: Does AI recommend tools that actually fit that stage? Different from pillar: Pillar prompts don't stack persona+stage; this is the variant that surfaces stage-specific gaps.
Long-tail discovery (2 prompts) — buyer-pain language
These mirror how buyers actually phrase queries when they don't know the category vocabulary yet.
Prompt 11 — Pain-anchored discovery
I'm struggling with [specific pain point]. What [category] tool can help me solve this?
When to run it: You want to test whether your brand surfaces for problem-led queries (Principle 2) at the long-tail end. Signal: Does AI translate the pain to your category and surface you? Different from pillar: The pillar's 2.5 ("tools that help with [problem]") is similar but doesn't include the first-person "struggling with" framing — a real buyer voice marker.
Prompt 12 — Switch-from-legacy framing
If I'm switching from [legacy approach: spreadsheets, manual process, competitor], what [category] should I consider that handles [migration concern]?
When to run it: You target buyers who are migrating from spreadsheets, manual processes, or a legacy competitor. Signal: Does AI surface you as a destination? With what migration framing? Different from pillar: Captures the migration intent specifically — high-conviction buyers who already know they need to switch. Zero pillar prompts test this.
How to customize templates for your category
The 12 above are the universal starter library. Your specific category may need additional variants. Three rules for writing your own:
- Anchor every variable in real buyer language. If you're writing a prompt with
[specific pain point], pull the pain language from a Reddit thread or sales call transcript, not from your messaging doc. - Test each new prompt 3 times before adding it to your tracked set. If AI's responses vary wildly across 3 runs, the prompt is ambiguous — refine the variables before locking it in.
- Document the diagnostic intent. For every prompt you add, write one sentence on what failure mode it diagnoses. If you can't articulate the diagnostic intent, the prompt isn't earning its slot in the audit.
The 60-minute exercise to build your custom set
| Step | Time | Source | Output |
|---|---|---|---|
| 1 | 20 min | Read 10 sales call transcripts (or talk to 3 AEs) | ~15 buyer-language questions |
| 2 | 20 min | Scan top 30 thread titles in your category subreddit | ~20 question-shaped prompts |
| 3 | 10 min | Pull first messages from your last 50 support tickets | ~15 information-gap prompts |
| 4 | 10 min | Adapt 4-6 of the 12 templates above to your category vocabulary | ~6 use-case variants |
After 60 minutes you have ~55 candidates. Apply the 5 principles as filters: cut anything written in your marketing voice, convert category-only candidates to problem-led variants, force at least 60% non-branded, and ensure the conversational form matches what buyers actually type. What survives is your tracked audit set. Lock it. Run it across ChatGPT, Claude, and Perplexity alongside the pillar's 15 starter prompts.
Where to run these prompts
The audit prompts in this post are designed to layer on top of the 15 starter prompts, not replace them. Run both sets in parallel:
- Pillar's 15 starter prompts — your universal baseline, re-run quarterly so you have stable run-over-run comparisons
- Spoke B's 12 use-case templates — your deeper diagnostic, run when you need to drill into a specific signal (sentiment drift, competitive shift, ICP gap)
The full multi-platform execution workflow is in how to track brand mentions in ChatGPT, Claude & Perplexity. The buyer-language mining playbook is in find buyer questions for AI prompts.
Frequently Asked Questions
How many AI visibility prompts should I track?
Start with the pillar's 15 universal starter prompts. Add 5-8 use-case variants from this post once you've identified which audit layer is your weakest. Above 30 total, the diagnostic value plateaus and the time cost becomes a tax on the workflow. The sweet spot for most brands is 20-25 prompts: 15 universal + 5-10 use-case-specific.
Should I write prompts in the first or third person?
First-person framing ("Should I use X for my team?") tends to elicit more advisory, opinionated AI answers. Third-person framing ("Is X good for small teams?") tends to elicit more list-style answers. Test in the form your actual buyers use. If you don't know which form they use, run both for a few prompts and look at which response shape matches what your sales team hears in calls.
How often should I refresh my custom prompt set?
The set itself should rarely change once locked — you need run-over-run comparison value. Refresh quarterly only if your category vocabulary has meaningfully shifted (new competitor, new use case, new ICP). Otherwise, the use-case variants stay stable. The pillar's 15 starter prompts essentially never change.
What if my buyers don't use AI search yet?
Audit anyway. AI surfaces are a leading indicator — buyers who don't use ChatGPT for evaluation today are typically using it within 6-12 months as it becomes embedded in workflows they already use (Notion AI, Google Workspace AI, Slack AI). The audit you run today is positioning for next year's buyer behavior, not just this quarter's.
Can I use the same use-case templates across multiple products or business units?
No. Adapt each template per product or per ICP. ChatGPT might know HubSpot the company perfectly and be vague on HubSpot Marketing Hub specifically. Same applies to multi-ICP brands. Buyer language and AI's recommendation patterns shift completely between segments, so each segment needs its own set with its own buyer-language inputs.
What's the difference between branded and non-branded prompts?
Branded prompts include your company name (e.g., "How does HubSpot compare to Salesforce?"). Non-branded prompts do not (e.g., "What are the best CRM tools for small SaaS teams?"). Branded prompts test validation and reputation. Non-branded prompts test discovery, which is where most large-volume buyer queries actually happen. A balanced audit needs both, weighted roughly 60-40 toward non-branded.
What if my sales team doesn't have time to provide buyer questions?
Spend 30 minutes reading the most recent 20 thread titles in your category subreddit. The titles ARE the prompts; people type their problems literally as they think them. This is the single fastest substitute for sales-call mining and produces 70% of the same value in a fraction of the time.
How do I know if a use-case variant is worth adding to my tracked set?
Three criteria: (1) it surfaces a different leaderboard from the pillar's universal version (cross-check by running both), (2) it produces consistent results across 3 runs (high-variance prompts are noise), and (3) you can articulate one failure mode it diagnoses. If a prompt fails any of those three checks, drop it.
About the author. Joao da Silva is co-founder of friction AI alongside Camilla Wirth. friction AI tracks brand visibility across ChatGPT, Claude, Perplexity, and Gemini for SaaS and DTC brands. Joao writes about AI search, entity recognition, and the operational side of getting recommended by LLMs. Connect with him on LinkedIn.