The best SaaS AI prompts paste real numbers before they ask for words. Positioning lines, onboarding emails, and churn interview scripts fail in predictable ways, and in each case the cause is the same: the prompt described a category of SaaS company instead of your company, so the model returned the average version of every idea in the category.
Below are three copy-ready prompts for the three moments where SaaS teams lose the most value from their own data: the position they sell from, the trial-to-paid email, and the cancellation conversation. Use them in that order, because the positioning decision constrains what onboarding can promise and what the churn interview should test.
Supply the numbers before you ask for the words
You are writing for [product name], a [what it is] used by [specific customer, e.g. operations managers at 50 to 200 person logistics firms]. Our verified numbers: price is [price and currency]; the plan limit is [limit]; setup takes [time]; the average customer reaches [the event that predicts retention, e.g. their third scheduled export] within [timeframe]; the most common churn trigger in our support log is [cancellation reason]. Write three positioning statements. Each is one sentence, each is under [character or word limit]. Each must name the customer, the job they are doing, and the mechanism that makes it possible. Do not use the words [banned terms your category abuses, e.g. innovative, next-generation, frictionless]. Do not invent a customer count, a percentage, or a competitor comparison. If a claim needs a number I did not give you, write [CONFIRM NUMBER] and list every marker at the end. Then name the single strongest objection each line invites, plus one sentence answering it.
Fill in these details
- Product name and category: the plain description a customer would recognise, not your internal codename.
- Customer: narrow enough that you can picture their week. A segment you can name is a segment you can sell to.
- Verified numbers: only figures you would defend in a procurement call. Leave a marker for anything you do not have.
- Retention event: the action that separates customers who stay from customers who leave in your own data.
- Banned terms: the words your category overuses. Naming them works more reliably than asking for a tone.
Delete the objection step for a landing page test and keep it for an enterprise pitch. If you sell to two segments, run the prompt twice with different customers and compare which line survives scrutiny.
Generate positioning options you can defend on a sales call
I sell [product] to [buyer role] at [company type and size]. Here is the exact sentence they say when they push back: [paste the objection in their words]. Here is what we can prove: [three provable claims, each with its source]. Here is what we cannot prove yet: [list the gaps]. Write four positioning lines, one sentence each, aimed at four different objections from that list. One line must be built on the strongest number we have. Do not use [banned words]. Do not promise an outcome we have not measured. After the four lines, tell me which one to test first and what evidence would show it failing.
Four lines rather than one gives you a decision instead of a decoration. Asking for the failure evidence keeps the test honest, because a positioning line that cannot fail is not a position.
Draft the onboarding email that matches what the trial promised
The trial email is where a positioning promise gets cashed or quietly broken. If your positioning line says setup takes ten minutes, no email should describe a twenty-minute setup.
Write a [number, e.g. four]-email onboarding sequence for users who started a [trial length] trial of [product] and have not yet [activation event]. Their current state: [what they have done so far, e.g. connected the account but run no export]. The promise my positioning line makes is [paste the positioning line]. The blocker at this step is [the real reason users stall]. Our product facts: [facts only]. Email [n] goes to [segment]. Each email is under [word limit] words and has one subject line, one link, and one action. Open with the reader's unfinished job, not with a greeting. Never mention a feature that is not in [the plan they are on]. Do not invent usage data, streak counts, or social proof. If an email needs a statistic, write [CONFIRM NUMBER]. End the sequence with an email that names what happens to their data if they do nothing.
Match the sequence length to time-to-value. If users reach activation in a day, four emails is already too many.
Build a churn interview script from real cancellation reasons
Here are the cancellation reasons customers gave us, most common first: [paste the list verbatim from your cancellation flow or support log, e.g. too expensive, not using it enough, switched tools, team changed]. Our plan price is [price]. The activation event is [event]. The average customer takes [time] to reach it. Write a twelve-minute interview script for a former customer who left in the last [time window]. Ask one question at a time and wait for my answer before the next. Do not defend the product. Do not offer a discount. Do not ask leading questions such as "was it the price". For each answer, ask at least one follow-up that asks for the specific moment rather than the general feeling. Then mark each answer as [activation never happened, activation happened but value stopped, or value was never proven], and flag anything ambiguous.
The one-question-at-a-time instruction is what makes this usable. Without it you get twelve questions to read aloud, which kills the interview. The three buckets turn a conversation into a product decision.
Worked example: repositioning a reporting tool
A small team sells a scheduled reporting tool at 39 dollars a month to operations managers. Their churn log's top reason is "not using it enough". Their activation event is a second scheduled export.
| Element | Average output | Filled output |
|---|---|---|
| Customer named | "operations teams" | "operations managers who send a weekly numbers email by hand" |
| Mechanism | "saves time with automation" | "the report builds itself from the export you already scheduled" |
| Number used | none | the price and the second-export activation step |
| Refusal rule | none | no invented counts, write [CONFIRM NUMBER] instead |
The filled version is not more creative. It is more falsifiable. Each line names something a competitor could not copy and something a sales call could check.
Check the output before it reaches a customer
1. Trace every number back to a source you control
Take each figure in the draft and find where it came from. If it came from the prompt, it is verified. If it came from the model's memory, delete it. Price, plan limits, and time-to-value are the three claims most likely to be wrong and most likely to be noticed.
2. Read each line as a customer who is already suspicious
Ask the question a careful buyer asks: how do you know? If the sentence cannot survive that question, it is copy, not a position, and it will fail in the first call.
Common SaaS prompt mistakes
- Asking for copy without pasting the numbers. The model fills the gap with category averages, and your positioning becomes interchangeable with every competitor's.
- Treating the AI as a customer. It cannot tell you which objection is real. Only your cancellation log and your sales calls can.
- Letting it invent social proof. One invented customer count in a landing page draft travels further than the rest of the page.
- Prompting once per asset. Positioning, onboarding, and churn answers contradict each other when they are generated separately. Generate them in sequence.
- Hiding the gaps. Listing what you cannot prove keeps the model inside your evidence, which is where marketing copy belongs.
Frequently asked questions
Can AI write positioning for a product that sells to enterprise buyers?
It can draft options, but enterprise buyers ask for numbers a marketing model cannot supply. Give it your security, compliance, and pricing facts, and keep your judgement on what to lead with.
How many onboarding emails should a trial sequence have?
Match the count to the time your users take to reach the activation event rather than to a rule of thumb. Fewer emails with one action each usually outperform a longer sequence repeating the same reminder.
Can an AI run a real customer interview?
No. It can generate the script and hold the follow-up discipline. The conversation, and the decision about what to change, still belong to you.
Run one prompt against your own numbers this week
Pick the stage where you are leaking the most, pull the real figures from your dashboard or cancellation log, and run the matching prompt once. Write [CONFIRM NUMBER] wherever the model wants a fact you do not have. For more founder-level prompts, browse the SaaS success prompt collection.