The operator's manual for prompting
A year into ChatGPT and Claude, you have hit the quality ceiling. Output drifts from one morning to the next. Prompts that work for you break the moment your team runs them. And when a vendor review asks where a number came from, you cannot answer.
Prompt & Profit is the operator's manual that closes the gap between using AI and operating it.
The first Playbook is built for marketing. The method behind it is not.
Proficient within an afternoon. Productive within a week. Expert within three months.
Two chapters free, no email gate. The full Playbook is $9.99/month — or $99/year, two months free.
7-day free trial · cancel any time
The evidence
Numbers carry their source and their grade — causal where a controlled study isolates the effect, survey where it is self-reported. The same standard the Playbook holds every claim to.
Marks belong to their owners; shown as citation, not endorsement.
The symptoms
If any of these were your last fortnight, you are who the Playbook was built for. They split two ways: operational friction that slows the work down, and governance gaps that cost you trust when it ships. Tap any problem to see the fix.
Day-to-day execution friction — the work that keeps not getting easier.
Same prompt, same model — materially different quality from one morning to the next. You lose a half-hour to the rerun, then an hour cleaning up. Good when it lands, a coin-toss the rest of the time.
How the Playbook fixes this: a dedicated post on why identical prompts still drift, and the determinism checks that catch it before publish.
A new flagship model lands every six weeks. Your team runs GPT for some jobs, Claude for others, and the choice is folklore. Vendors say use "the smartest model" — which maximises their bill, not your output. Pick wrong and you either overpay for a job a cheaper model nails, or ship weak work from a model that never fit it.
How the Playbook fixes this: the model-choice matrix laid out by job shape, with the swap test for new releases.
Your team's best prompts live in private chat histories, in screenshots pasted into Slack, in a Notion page someone built and abandoned. There is no single source of truth. So every new starter rebuilds the same prompts from scratch, and the craft you have already paid for never compounds.
How the Playbook fixes this: every prompt follows one YAML schema — declared inputs, a worked example, stated failure modes — so the library is one source of truth, not eight chat histories.
The prompts that fire for you misfire for everyone else. The library you spent a quarter building has half its prompts dead by the time a sister team picks it up. The knowledge never actually left your head.
How the Playbook fixes this: a post on why prompt libraries fail at hand-off, and the schema discipline that lets a second team run the work cold.
The prompt that worked last quarter is not the prompt that works today — you have tweaked it three times since. You cannot say which version produced the brief that landed, and you cannot roll back when a "small improvement" quietly degrades the output. There is no diff, no changelog, no undo for the one artefact your team's craft compounds in.
How the Playbook fixes this: a dedicated post on treating prompts as code — diff, revert, attribution per template — with the smallest discipline that gets you there.
The layer your CMO, legal team, and finance team see — defensibility, audit, measurement.
The model invents a Forrester benchmark, fabricates a journal title, pins a confident quote on a real analyst. The numbers look plausible, so you ship the post. A reader clicks the footnote; the URL 404s. You spend the next morning auditing which figures in the back catalogue are even real.
Current reasoning models hallucinated on 33% of PersonQA prompts — a smaller sibling on 48% (OpenAI o3 / o4-mini system card, 2025). Fact-bearing output without a verification gate ships those rates to your readers.
How the Playbook fixes this: the evidence standards set out the source tiers and the verification rule that keeps fabricated citations from getting past draft.
The CMO asks where the bolded statistic came from, and the trail goes cold. Legal flags the AI-generated copy in a vendor review. The work already shipped — and you have no audit trail to stand behind it.
How the Playbook fixes this: a dedicated post on the audit trail every AI-generated draft needs — source, tier, verification, authorship — before publish, not after.
The team is shipping more — everyone can feel it. Finance wants it in numbers they will accept: drafts per writer per week, time-to-publish, cost per qualified piece. You can feel the gain; you cannot put it on a slide.
93% of CMOs claim measurable ROI from generative AI, yet Gartner projects 30% of generative-AI projects are abandoned after proof-of-concept (SAS & Coleman Parkes, 2025; Gartner). The gap is measurement.
How the Playbook fixes this: the measurement framework — drafts per writer per week, time-to-publish, cost per qualified piece — ships in the Best Practice Guides bundled with the subscription.
The system
Every problem above has one root: you are prompting from memory. The output drifts, the craft does not compound, and you cannot defend the result. The Playbook replaces the improvisation with three things you can hold, hand off, and audit.
Every template declares its inputs, carries a worked example, and states its failure modes — so a second person runs it cold and the library stays one source of truth, not eight chat histories.
Source tiers and a verification rule stop fabricated citations at draft, not after a reader clicks a footnote that 404s.
Marketing is one of the few functions where the gain is directly countable — drafts per writer per week, time-to-publish, cost per qualified piece. The Playbook wires those metrics in, so "we ship more" becomes a figure your finance team will sign off.
Run the work this way and the gains stop being anecdotal. The cases below are what a disciplined library produces when the method holds.
Named results
Two named cases from the verified register, each with its source and its grade. Vendor-published figures are ceilings, not medians — your numbers move with your data and the task. No anonymous "major retailer" claims, by rule.
~11 years of work, in months
CarMax with Microsoft · Azure OpenAI
Over 100,000 customer reviews synthesised into 5,000 vehicle research pages, at an 80% editorial approval rate.
+82% conversion rate
HubSpot on its own CRM stack
AI-assisted personalisation of a lead-nurture email flow — with +30% open rate and +50% click-through.
The three stages
Three stages, three clear outcomes. Open a stage to see what changes for you and how you get there — free at the start, paid where the work goes deeper, direct when you need a second opinion.
Stage 1 of 3 · Within an afternoon · Proficient
The arc from "I have used ChatGPT" to "I run deliberate prompts" closes in one sitting. You finish the afternoon able to take a brief into a model and come back with output you would actually ship.
How you get there: two complete sample chapters and the insights, both free in your browser. No subscription, no email gate.
Stage 2 of 3 · Within a week · Productive
You have a prompt library with declared variables, worked examples, and stated failure modes. Your team can pick up any template and run it without you in the room. The measurement spine — drafts per writer per week, time-to-publish, cost per qualified piece — is wired up.
How you get there: the Digital Playbook subscription. Fifteen chapters, sixty-plus prompt templates, the verified case-study register, every Best Practice Guide as it ships. $9.99 / month solo · $45 / month team of five · bespoke beyond.
7-day free trial · cancel any time · billed monthly or yearly (two months free on annual).
Causal The return you are pricing against: controlled trials put the gain at 40% faster drafting and 25% faster knowledge work (Noy & Zhang, Science 2023; Dell’Acqua et al., Harvard / BCG 2023).
Stage 3 of 3 · Within three months · Expert
You train new starters on the Playbook standards. You design review checkpoints that clear a legal pass. You answer vendor-decision questions in the call rather than after it. The Playbook becomes the second-nature reference your team works against.
How you get there: direct work with the author. Advisory calls — $249 for a sixty-minute call — when you need a second opinion on a specific question; Project Implementation Consulting, scoped by proposal, when you need help shipping a project end to end.
The Playbook line
The Playbook you subscribe to today is built for the marketing function and the skillsets around it. It is the first in a line that grows along four axes. The same method carries across all of them — deliberate prompts, a versioned template library, and the evidence and case-study standards you can already audit. What changes is the cut, not the discipline.
Today, your subscription is the marketing Playbook. Each new one ships only when it clears the same bar this one did — no dates promised, no placeholder chapters. When a Playbook lands, it lands complete.
The standards
Trust is built through restraint. The Playbook's standards are public so you can hold the work to them.
Source tiers, citation rules, and the verification every statistic clears.
Read the standard →The YAML schema every prompt follows — declared inputs, worked example, failure modes.
See the schema →The bar every named-company case clears before it enters a chapter.
Read the standard →Objections
The objections worth answering before you read a line of the sample.
No. Searching gives you screenshots and folklore that do not survive a hand-off to your team. The Playbook gives you a versioned prompt library on one schema, evidence standards you can audit, and a measurement spine your finance team will accept.
It is model-agnostic. Instead of betting on one vendor, you get a model-choice matrix by job shape and a swap test for when a new release lands — so the method outlives the model of the month.
Yes. Every subscription opens with a 7-day free trial, cancels any time, and bills monthly or yearly — two months free on annual.
Yes. You read the sample in your browser with no email gate and no card, so you can judge the work before you subscribe.
No. If you can write a brief, you can run the Playbook. There is no code to write — the discipline is editorial, not engineering.
The Playbooks are living artefacts. If you spot an unsourced statistic, a citation that does not resolve, or a case that has materially changed shape — flag it.
Substantive corrections land in the v1.x update track within sixty days of report. See the update log for what shipped when.