SolCrys Logo

How SolCrys Works

How we build prompt sets - the Golden Prompt Set methodology

We build a Golden Prompt Set (GPS) for every workspace from real demand signals for its category: search demand, the questions people ask around it, and community discussion where available. Every prompt stands for a buyer intent, named by its prompt type, and for the persona asking. The set is balanced across types, each of which tests a different way your brand can be invisible, and prompts carry a demand band so you can prioritize. We publish how we choose because prompt selection is the single biggest determinant of AEO data quality. If we're tracking the wrong prompts, every chart in your dashboard is wrong, so we'd rather you evaluate how we choose before you sign than after.

Published · Updated

Questions this guide answers

  • How does SolCrys choose which prompts to track?
  • What is the Golden Prompt Set?
  • How does SolCrys decide which prompts to monitor for my brand?
  • Where do SolCrys's prompt sets come from?

Direct answer

We build a Golden Prompt Set (GPS) for every workspace from real demand signals for its category: search demand, the questions people ask around it, and community discussion where available. Prompts carry a demand band so you can prioritize, and the keyword behind them so you can see why they're in your set.

Buyers word the same need in endless ways, but the intent behind it and the persona asking stay stable. A GPS is built around those: each prompt stands for an intent and a persona, and the set reruns the same way every time, so a change in the numbers means something.

We treat prompt selection as the single biggest determinant of AEO data quality. If we track the wrong prompts, every chart in your dashboard is wrong - so the rest of this page explains how we choose, what we deliberately exclude, and how you can audit our selection.

Three failure modes we built the GPS to avoid

We've watched AEO prompt selection go wrong in three specific ways. The GPS exists to avoid all three.

Failure modeWhat goes wrongWhat buyers experience
Reformatted keyword listsTools pull SEO keywords and reformat them as questions ('best [X]?', '[X] vs [Y]?'). They look like prompts but don't match how buyers phrase questions to AI assistants.'Why is my AEO tool tracking 200 prompts but I never see traffic move when I fix things?'
Customer-only promptsTools rely entirely on customer-supplied prompts. Marketers tend to ask brand-flattering questions; real buyers ask harder, more skeptical ones.'Our SOV looks great but our pipeline isn't growing.'
Category-only promptsTools generate '100 questions a buyer might ask' from the category name alone, with no demand data behind them. The list sounds plausible, but nothing ties it to what buyers search for or discuss.'We're tracking lots of prompts but only 3 of them seem to drive any decisions.'

What a GPS is built from

Every GPS starts from demand signals for your category. The exact inputs, suppliers, and generation process are implementation details we keep confidential; what they cover is below.

Search demand

Keyword demand for your category in search, and in marketplace search for categories where buyers shop there.

Intent volume is the closest public proxy to real demand. A query asked at high volume in conventional search is highly likely to be asked in similar form in AI assistants. The phrasing differs, since AI prompts are longer and more conversational, but the underlying intent is the same.

The questions people ask around it

Buyers put full, problem-stated questions to AI assistants, not head terms. That gap between a keyword list and a prompt set is why the questions people ask around a keyword matter as much as its volume.

Community discussion, where available

Community discussion adds how buyers phrase questions to other people, which is close to how they ask AI.

What a Golden Prompt Set looks like in practice

A GPS build usually produces more candidate prompts than a workspace actively tracks. The tracked benchmark should match the customer's plan and scope: current public brand plans track 10 prompts on Free, 20 on Starter, and 60 on Pro; Agency plans track 30 prompts per client organization; Enterprise scope is confirmed during evaluation.

The mix is not improvised per customer. Every prompt type answers a different visibility question you can fail on independently, and each set is built to a fixed type mix, not padded with near-duplicate variants. There are two mixes. Discovery focus is the default for onboarding, the Free AI Visibility Audit, and Suggest prompts: 60% Discovery, 30% Feature Research, 10% Brand Accuracy, and no Comparison prompts. Balanced, which you can choose when you generate prompts, is the mix in the table below. Examples use well-known public products purely for illustration - they are not taken from any customer's prompt set.

Discovery focus leaves Comparison out on purpose. A Comparison prompt names your brand inside the prompt itself, so the engine has no way to omit you; it measures how you're framed, not whether you're found. The Brand Accuracy prompt names your brand too, and it tests whether the facts are right. Read those two types separately from Discovery and Feature Research, the prompts that test unaided discovery.

We also deliberately do not generate 'alternatives to [competitor]' prompts. Anchoring a prompt on a single competitor's name measures that competitor's gravity more than your category presence, so we rewrite that intent as brand-neutral Discovery instead.

Prompt typeShare in the Balanced mixThe visibility question it testsIllustrative example
Discovery50%Does the engine name you for the broad category question, when no brand is named in the prompt?Best CRM for a 50-person B2B SaaS team
Feature Research30%Does the engine name you for a specific persona, use case, capability, or constraint?Which CRM handles a 6-month enterprise sales cycle?
Comparison10%When a buyer puts you head-to-head against a named competitor, how are you framed?How does HubSpot compare to Salesforce for a 50-person team?
Brand Accuracy10%When a buyer asks about you directly, are the facts and the sentiment right?Is [your brand] worth the price?

The metadata we attach to your prompts

Your prompts stay private to your workspace. For the prompts in your GPS we attach metadata so you can filter, sort, and prioritize - and so you can audit how the prompt got there.

  • Demand band - High, Medium, or Low, from the demand behind the prompt's keyword, so you can prioritize the questions that carry the most buyer interest. Brand Accuracy prompts test facts, not demand, so they run without one.
  • Prompt type - which of the four visibility questions above the prompt tests. The types are failure-mode oriented: each one is a distinct way your brand can be invisible, and you can fail one while passing the others. Type is the field our distribution targets are enforced against - we don't layer a separate journey-stage label on top of it.
  • Target persona - the buyer role the prompt is written for, such as a VP of marketing or a data center architect. Every prompt carries one, and when no specific role is cleanly inferable we set it to Generic rather than inventing a persona to look thorough. Persona is a filter and a sort column in your Prompts table and a column in your CSV export.
  • Per-engine results - once the prompt runs, each response is stored per engine, so you can see which engines surface you for that prompt and which don't. This is measured after the fact, not predicted: we don't guess engine fit at authoring time.
  • Measurement anchor - the keyword behind the prompt, shown in the demand band's popover, so you can see why it's in your set.

How your own prompts blend in

We treat the GPS as the foundation, not the whole thing. You know your buyers in ways we don't, so every workspace lets you supply your own prompts as an alternative to, or alongside, the GPS. We default to running the GPS so teams without time to write prompts still start from a demand-based set. Questions from your own sales calls, support tickets, and demos are the closest thing to real buyer prompts any team has, and they belong in the set. In ongoing workspaces, you can replace any GPS prompt with your own at any time, or run Suggest prompts from keywords or a page URL to generate new candidates you approve.

The prompt sanity-check workflow is coming soon. Today, we treat off-topic prompts, brand-flattering prompts without comparative context, and duplicate prompts as review criteria during setup and customer review. When this ships in-product, flags will be guidance rather than gating.

How we keep the GPS stable

The GPS is designed to be a stable measurement benchmark, not a real-time prompt rewrite engine. Prompt changes are handled as review items with the customer during onboarding, regular measurement reviews, customer-requested replacements, or clear category changes such as a new entrant, pricing shift, or major news event.

Review momentWhat changes
Onboarding / workspace setupPrompt fit is reviewed with the customer before the benchmark is established.
Regular measurement reviewWe revisit prompt fit, commonly monthly, without changing tracked prompts automatically.
Customer-requested replacementCustomers can replace prompts; unchanged prompts keep historical trend lines.
Category changeMajor news, new entrant, pricing change, or category spike can trigger a suggestion review.

What we are explicit about not promising

We've written this section because we'd rather you know the limits up front than discover them in your second quarter with us.

  • We don't promise the GPS guarantees revenue lift. A perfect prompt set doesn't fix weak content or thin brand presence; it just measures the right things.
  • We don't claim to replace human judgment. We support your own prompts because you know your buyers in ways we don't.
  • We don't automatically change tracked prompts without approval. Prompt revisions are handled as review items with your team; newly added prompts start their own measurement history.

Why we publish this methodology

Most platforms in our category keep prompt-selection methodology opaque. We publish how ours chooses because trust is our product - if you don't trust how we chose your prompts, every chart we show you is meaningless. We want this methodology to be falsifiable: open any demand band to see the keyword behind it. We want you to evaluate us before you sign, not after.

FAQ

How is the Golden Prompt Set different from just using SEO keywords?

SEO keywords measure what people type into a search box. AI prompts are longer and more conversational. In a GPS, keyword demand is evidence, and each prompt reads like a question a buyer would put to an AI assistant rather than a reformatted keyword.

How many prompts should I be tracking?

Start with the tracked prompt budget included in your plan, then keep the set stable enough to measure over time. Current public brand plans track 10 prompts on Free, 20 on Starter, and 60 on Pro. Agency plans track 30 prompts per client organization, and Enterprise scope is custom. The GPS framework can generate additional candidate prompts, but the prompts you actively track are the benchmark.

Can I track prompts in languages other than English?

Currently the GPS supports English-language prompts across English-dominant markets (US, UK, Canada, Australia). Multi-language GPS support is on the roadmap; please contact sales for current localization availability.

Do you share my custom prompts with anyone?

No. Customer-supplied prompts are private to your workspace and never used to build other customers' prompt sets.

Are GPS prompts based on real buyer demand?

Yes. Every GPS starts from real demand signals for your category: search demand, the questions people ask around it, and community discussion where available. The prompts represent the buyer intents that carry demand, written the way a buyer would ask an AI assistant, and a prompt's demand band shows the keyword behind it. Add questions from your own sales calls, support tickets, and demos alongside them.

What happens to my historical data if my prompt set changes?

Your tracked prompts don't change unless you approve the change. Prompt revisions are handled as review items with your team; historical trend lines for prompts you keep tracking stay continuous, while newly added prompts start their own measurement history.

Free ChatGPT visibility check

See where AI answers skip your brand — then fix it, free

Start a free workspace with your domain: 10 buyer-intent prompts through ChatGPT show where you are mentioned, cited, or skipped, and who gets recommended instead. A free content audit in the same workspace hands you the first fix to ship.

Start Free

Free · No credit card · About 5 minutes

Related guides

How SolCrys Works

How to Set Up a SolCrys Workspace: A Best-Practices Guide

What a SolCrys workspace is, the two ways to create one, and the best practices that keep your AI-visibility measurement clean: scoping the category, when to use multiple workspaces, engines, competitors, Corporate Context, and the prompt set.

How SolCrys Works

SolCrys FAQ

How we build prompt sets, which engines we cover, how we handle engine non-determinism, what happens after the measurement, what it costs, and how we handle your data - the 33 questions buyers ask during evaluation.

How SolCrys Works

AI Visibility Measurement Methodology

How we capture your AI visibility data across supported engines, with each response traceable to a prompt, engine, capture method, available model or surface signal, and timestamp. Consumer-surface and retail-assistant validation are scoped where technically reliable.

Buyer Guides

Evaluate an AEO Platform's Data Methodology

Six questions every buyer should send to every AEO platform - including us - before signing. We designed SolCrys to answer all six; here's how, and what to listen for from anyone you're evaluating.

Prompt Intelligence

AI Search Prompt Set

A practical guide to building an AI search prompt set across category, comparison, risk, implementation, competitor, and brand-specific prompts.

Measurement

AI Share of Recommendation

AI Share of Recommendation measures how often answer engines recommend a brand, not just whether they mention it. Learn how to track and improve it.