Mention Clock

How to pick the 20 prompts that matter

Twenty is a reading limit, not a magic sample. Keep the questions a buyer actually asked, and drop the ones you cannot act on.

Built by StyleLine FZE in Sharjah. Email support@mentionclock.com. We reply within one business day.

A SaaS brand can brainstorm 80 prompts in an afternoon and learn nothing, because nobody will reread 80 answers. Twenty is a cap you can still review: enough to cover category, job, switching, constraint, and buyer role, and small enough that a lost prompt has an owner. It is not a research finding about how many questions "the market" asks. Mention Clock does not publish a benchmark for a correct list length.

The cap also matches a real plan. Starter is 25 prompts, so a list of 20 leaves 5 slots for a launch question without a plan change. Free is 5 prompts, so the first month you run the 5 of the 20 that sales hears most often. Pro is 100 prompts if you later expand. Do not fill unused slots with slogan rewrites. Empty slots do not change the invoice.

AI prompt ideas by industry is the brainstorm. Prompt tracking for brands is the weekly habit once the wording is frozen. The job here is the cut.

Collect questions you can source

Start from sentences a buyer already used. Pull them from lost-deal notes, trial cancellation reasons, sales calls, and support threads where someone asked "what should we use." Keep the job, the role, and the constraint. Drop your slogan.

Add the jobs the product actually does, written the way a buyer would ask before they know your name. If the product sends invoice reminders, the question is about reminders, not about "a modern platform."

Skip keyword tools as a source of volumes. We do not have search volumes for chat prompts. A made-up volume would make the shortlist look more scientific than it is. A question that showed up twice in real calls outranks a question that only appeared in a brainstorm.

Write each candidate on one line, with three marks beside it:

  • Heard: a buyer said something like this, or they did not.
  • Decisive: a wrong answer would change a deal, or it would not.
  • Actionable: if you lose, there is a page or a listing you can edit, or there is not.

A prompt needs two of the three marks to survive the first cut. A clever question nobody asked, that you also cannot answer with a page, is how lists get long.

Five buckets for a SaaS brand

Keep the buckets uneven if the calls are uneven. A security-led product may end with more constraint prompts than category prompts. Forcing five equal groups hides the buying motion you actually have.

Category. The buyer names the shelf. "What warehouse slotting software should a 40 person 3PL shortlist?" You either appear on that shelf or you do not.

Job, before the category. The buyer names the pain. "How do we stop double-booked dock appointments without hiring another planner?" You can show up before they learn the category name.

Switching. The buyer already has a vendor. "What are alternatives to [rival] for teams that outgrew spreadsheets?" Use a rival your sales team actually loses to. Do not invent a competitor you have never seen in a deal.

Constraint. Platform, identity, data residency, or a workflow limit. "Which billing APIs support usage records and a Salesforce sync?" These prompts are where a generic "best of" list goes quiet and a specific product can be named.

Role. Founder, admin, finance, or the person who will live in the tool. "What should a finance lead check before buying usage based billing?" Role prompts catch the second reader on a deal, the one who was not on the first call.

A planning sketch you can copy and then replace

The lines below are shapes for a fictional usage based billing API. They are a worksheet, not a customer log and not a set of recorded answers. Replace the job, the rival names, and the constraints with yours before you track anything.

Category:

  1. Best usage based billing tools for a B2B SaaS with annual contracts and overages.
  2. Which billing platforms can invoice mid cycle when usage spikes?
  3. What billing software do product led companies use once Stripe invoices are not enough?
  4. Shortlist billing APIs for a team that already has a product catalog in Stripe.

Job:

  1. How do we bill customers for metered API calls without building an invoicing service?
  2. How should a SaaS handle prepaid credits that expire?
  3. What breaks when usage billing is done in a spreadsheet at renewal time?
  4. How do finance teams reconcile usage invoices with the product database?

Switching:

  1. Alternatives to a homegrown usage script once the company has more than one product.
  2. What should we look at if our current billing vendor cannot do credits?
  3. Migration risks when moving usage records from one billing system to another.
  4. Which teams replace manual overage spreadsheets, and with what?

Constraint:

  1. Billing APIs that accept usage records and post to Salesforce.
  2. Usage billing that can show a customer their running total before the invoice.
  3. Can a billing tool keep tax handling outside the usage meter?
  4. What do we need if some customers are invoiced and some pay by card?

Role:

  1. What should a founder check in a billing vendor before a Series B finance hire arrives?
  2. Questions a controller asks about usage invoices and revenue recognition.
  3. What a support lead needs so agents can explain a usage charge.
  4. How a product manager should describe meters so sales does not invent a new one per deal.

Notice what is missing. There is no "Why is [your brand] the best billing API?" A prompt that already names you measures whether the description is accurate. Keep at most one or two of those, and keep them outside the 20 if you are trying to see whether you are recommended when the buyer does not know you. Accuracy checks are a different job. What does AI say about my brand is that job.

Cut to 20 without a debate

Read the candidate list in a 45 minute meeting with one person from sales and one person who can edit the site. For each line, confirm the three marks. If you have more than 20 survivors, drop in this order:

Drop prompts that mention only your brand.

Drop prompts whose only difference is a synonym. "Best usage billing" and "top usage billing platforms" are one question written twice. Tracking both spends two slots to learn one thing.

Drop prompts you cannot act on. If a loss would require a feature you will not ship, the prompt is a wish, not a monitoring task.

Drop prompts tied to a single logo you do not lose deals to. A famous rival in a blog post is not automatically a rival in your pipeline.

Stop at 20 even if a few good lines remain. Park them in a backlog note. The point of the cap is that the weekly review stays readable. Share of voice in AI search only means something if the rival set and the prompt list stay still. Adding a 21st prompt next Tuesday starts a new survey.

Freeze the wording

Write the final 20 in a doc the whole team can see. Include the role, the job, and the constraint inside the sentence. Do not leave those in a separate "notes" column that the runner forgets to paste.

Rules that keep next month comparable:

Do not put your brand name in the category, job, switching, or constraint prompts. A named brand prompt answers a different question.

Name a rival only in the switching bucket, and spell the rival the way buyers spell it.

Do not "improve" a prompt because last week's answer was annoying. A changed sentence is a new series. If the wording was wrong, retire it and say so.

Keep one owner who can reject edits. Sales will want to add a prompt after every lost deal. The owner adds it only by removing one, until you move to a plan with a higher cap.

Fit the 20 to a plan

Free is $0, 1 project, 5 prompts, 2 engines, weekly, no card. Pick the 5 prompts sales can repeat from memory. Run those until the review is a habit.

Starter lists at $29 each month, and at $290 billed yearly: 1 project, 25 prompts, 3 engines, weekly, 3 rivals, Slack or Teams. The full 20 fits, with 5 slots unused.

Pro is $79 a month or $790 a year: 100 prompts, every 3 days, 10 rivals, and a cited source report. Use it when 20 is no longer the reading limit you chose, not merely because a plan name sounds larger.

The engines on a paid plan today are ChatGPT, the second chat engine, and Perplexity. Free runs two of those. The upcoming engine is being added. Do not write the upcoming engine into the prompt doc as if the scheduled run already includes it. A hand check on another surface stays labeled as a hand check.

Each prompt runs on every engine in the plan. Twenty prompts on Starter's three engines is 60 answers a week. That is the reading load. If nobody will open 60 rows, you picked a plan that is wider than the review, and you should cut engines or cut prompts rather than skim.

What you do with a miss

A miss means the answer did not name you, or it named you with a wrong fact, or it cited a URL you do not control. Assign one URL. Ship or request one edit. Leave the other 19 prompts alone that week.

If the miss is "a rival was recommended and we were not," the follow up is content and listing work, not a new prompt. That sequence is what to do when ChatGPT recommends a rival. Do not respond by adding five more ways to ask the same question.

When the 20 are written down, put the first 5 on a clock: create a free project. Move the rest onto Starter when you are ready to read 20. The index of the other writeups is on resources.

Related guides

FAQ

Why 20 prompts and not 5 or 100?

Five is what Free can hold, and it is a fair first month. One hundred is what Pro can hold, and most teams will not read it. Twenty fits inside Starter's 25 prompt cap and still covers category, job, switching, constraint, and role. It is a reading limit.

Should every prompt mention our brand?

No. Prompts that already name you test accuracy. Prompts that do not name you test whether you are recommended. Keep those jobs separate. Most of the 20 should omit your name.

Can I track the same 20 on ChatGPT, the second chat engine, and Perplexity?

Yes on Starter, Pro, and Agency, which include three engines. The live engines are ChatGPT, the second chat engine, and Perplexity. The upcoming engine is being added. Free includes two engines, so you choose two and you say which two in the readout.

What if sales wants to change the list every week?

They can propose a swap. The owner removes one prompt for each prompt added, until you raise the cap. A renamed prompt is not a continuation, and the share figure stops being a trend.

Create a free project