Intro
A baseline is only as good as the questions it asks.
The prompt set is the spine of every Capston Core engagement. It decides which queries we observe, which competitors get a fair test, and which gaps the brand can act on. Build it well and the score is comparable from quarter to quarter. Build it loosely and every retest is a new project.
This page is the operational guide: the four intent buckets, where to source prompts, how many are enough, and why the set has to be locked before captures begin.
Why the prompt set is the foundation
Every measurement that follows depends on the prompts.
The AI visibility scoring system aggregates eight dimensions — brand presence, answer position, citation share, source quality, competitor dominance, fact accuracy, sentiment, commercial risk. Each of those dimensions is computed against captured answers. Each captured answer started life as a prompt.
Change the prompts and you change the score. Change the score and you have nothing to compare to last quarter.
The prompt set is therefore the first deliverable of the Capston Core methodology and the artefact that anchors the entire engagement. It is co-designed in week one, signed off before any captures begin, and stored alongside the captured answers in the AI answer evidence layer so any number can be traced back to the prompt that produced it.
The four intent buckets
A usable prompt set covers the full buying journey, not just the moment someone types a brand name. We split prompts across four intent buckets.
Discovery
Broad queries where the brand is not yet in the conversation. The user is asking the AI to surface a shortlist.
- “best luxury resort in Mauritius for families”
- “where to stay in the Loire Valley for a wine weekend”
- “premium safari lodges in Botswana under USD 2,000 per night”
Discovery prompts test whether the brand makes the cut at all. They are the hardest to win and the most commercially meaningful.
Comparison
Head-to-head and “best of” queries. The user is short-listing.
- “Hotel A vs Hotel B for honeymoon”
- “best beach resorts in the Maldives compared”
- “Royal Palm vs Constance Le Prince Maurice”
Comparison prompts reveal which competitor the AI engines pair the brand with — and on which dimensions the brand loses (price, location, service, reputation).
Trust
Review-driven, reputation-driven, accuracy-driven queries. The user is checking before booking.
- “is [hotel] worth the price”
- “[hotel] reviews 2026”
- “is [hotel] family friendly”
Trust prompts surface what AI engines are saying about the brand when the user is already interested. This is where wrong facts and stale descriptors do the most damage.
Conversion
Branded queries where the AI answer drives or breaks the booking.
- “book [hotel] direct vs Booking.com”
- “[hotel] best rate”
- “[hotel] cancellation policy”
Conversion prompts are the commercial frontier. They expose OTA capture, aggregator routing, and intermediary substitution — the dimensions the Capston Hospitality Scorecard flags as commercial risk.
Sources to draw prompts from
A good prompt set is sourced, not invented. Five inputs feed it:
- Real query logs — site search, internal chat logs, support tickets. These are real customer questions in real customer language.
- GSC data — top queries by impressions and clicks. They reveal which discovery and trust questions the market is already asking.
- AI engine suggestions — “people also ask”, related questions, follow-up suggestions inside ChatGPT, Perplexity, and Google AI Overviews. The AI tells us what it expects users to ask next.
- Customer interviews — five to ten conversations with recent guests or buyers about how they researched the decision. Captures the comparison and trust language that never reaches a search box.
- Sales calls — the questions prospects actually ask the front desk, reservations, or sales. These are conversion prompts in their natural form.
Each source covers different intent. Query logs and GSC bias toward discovery and trust. Customer interviews surface comparison. Sales calls map conversion. Together they cover the four buckets without guesswork.
How many prompts are enough
Forty to eighty.
Below 40, the score is noisy: one capture per dimension is too thin to spot a pattern. Above 80, the workload of capture, QA, and retest grows faster than the insight it produces.
Within that range:
- 40–50 prompts — single brand, single market, focused scope. Most pilots land here.
- 50–65 prompts — single brand, two to three markets, or a brand with a strong comparison set.
- 65–80 prompts — portfolio brands, multi-market, or verticals with long buying journeys (luxury hospitality, B2B premium services).
Distribution across the four buckets is roughly balanced but not rigid: discovery and comparison typically take 50 to 60 percent of the set in hospitality, with trust and conversion sharing the rest. The mix is adjusted per brand and signed off with the rest of the prompt set.
Why the set must be locked
This is the rule that protects every number that follows.
Once the prompt set is signed off and captures begin, the set is locked. No additions, no removals, no edits between the baseline and the retest. If a stakeholder spots a missing prompt mid-cycle, it is logged and added to the next cycle — never retro-fitted.
The reason is simple: the score is a comparison. Baseline score versus retest score only means something if both ran the same prompts on the same engines with the same competitor set. Add a prompt mid-cycle and the comparison is gone. Remove one and the trend line bends for the wrong reason.
New prompts open a new cycle. The previous cycle stays intact, archived in the AI answer evidence layer, comparable to its own baseline. Brands that respect the lock get clean quarter-over-quarter movement. Brands that don’t, get noise.
How this fits into Capston Core
The prompt set is the first artefact produced in the Capston Core methodology and the dependency for everything that follows: captures, AI visibility scoring, the AI answer evidence layer, and the Capston QA standards that govern how the set is reviewed before sign-off.
Co-design happens in week one. Sign-off closes week one. Captures begin in week two. The set then stays locked until the next cycle opens.
→ Back to Capston Core
FAQ
How long does it take to build a prompt set?
Five to seven working days from kickoff: two days to source, two days to draft and bucket, one to two days for client review and sign-off.
Who signs off the prompt set?
The brand stakeholder responsible for the engagement — typically marketing director or commercial director. Sign-off is documented and stored with the captures.
Can the set be changed mid-cycle?
No. New prompts open a new cycle. The previous cycle stays locked so the baseline-versus-retest comparison remains valid.
Do all four buckets need equal weight?
No. The split is adjusted per brand and per vertical. Hospitality typically leans 50 to 60 percent on discovery and comparison; B2B premium services lean more on trust and conversion.
Final CTA block
Build the prompt set behind your AI visibility baseline.
Score your brand
Read the methodology