Product managers spend more time on textual content than so much other people recognise. Specs, user reviews, acceptance criteria, rationale notes, stakeholder updates, the extraordinary govt transient. The paintings is an element architecture, component translation, edge negotiation. The writing is wherein alignment gets developed or damaged.
ChatGPT can assist the following, no longer as a substitute for judgment, yet as a drafting spouse that speeds up the dull materials and sharpens the messy ones. Used well, it is going to keep hours in step with week and reduce remodel across engineering and design. Used poorly, it floods your backlog with primary fluff that no person reads. The big difference is in how you steered, whenever you carry it into the system, and how you overview what it produces.
I have deployed ChatGPT across discovery, planning, and supply on groups ranging from scrappy five-grownup startups to business courses with compliance gates and multi-region timelines. The styles in this newsletter come from that enjoy. They center of attention on specifications and person testimonies because the ones artifacts deliver so much of the paradox chance. Good prose right here is leverage.
Where ChatGPT matches inside the PM writing flow
Think of PM writing as three levels: exploration, crystallization, and verbal exchange. Exploration is the place you pull threads and bring together inputs. Crystallization is where you opt. Communication is wherein you get the group shifting. ChatGPT works most well known on the transitions between these levels.
During exploration, it enables you disintegrate a messy set of notes into subject matters, spot contradictions, and sketch one-of-a-kind answer frames. In crystallization, it could generate various spec outlines, propose popularity standards variations, and rigidity try assumptions with side circumstances. For verbal exchange, it turns choices and rationale into clear consumer reports and stakeholder summaries tailored for one-of-a-kind audiences.
The vital idea is to anchor the edition with concrete context. Feed it genuine person charges, precise improve tickets, rough tips, design snapshots, and a one-paragraph process summary. The output caliber rises dramatically while the kind is limited via the certainty of your product and clients.
Drafting crisp specs devoid of shedding nuance
A spec need to be a resolution document. It needs to explain what will substitute for the person, why that alternate topics, the limitations, and how we shall comprehend it labored. The seize with ChatGPT is verbosity. It will fortunately produce five pages the place one might do. Your job is to drive toward tightness.
Start with a one-paragraph hassle framing to your very own words. Keep it to a few sentences: who is affected, what is broken, and what happens when it’s mounted. Paste related snippets of proof: two user rates, a day-by-day question matter from analytics, a screenshot or redacted price ticket thread. Then ask ChatGPT for 2 or 3 replacement spec outlines, every under one web page, with the several industry-offs. You will by and large see a framing you had not thought-about.
I choose a spec architecture it truly is heavy on what and why, and gentle on implementation aspect. Engineering wants space to design. The spec ought to outline limitations, now not dictate code paths. ChatGPT can lend a hand by way of list disadvantages, metrics, and non-desires you would pass over, notably whilst you ask it to generate pushback. Tell it to argue towards your plan from the point of view of a senior engineer or a skeptical compliance officer. Often that yields a line or two that ameliorations the scope in important techniques.
Resist the urge to replicate the 1st coherent draft into your doc. Instead, sew in combination sentences from the major version, upload references for your actual artifacts, and trim anything vague. ChatGPT is an accelerant, no longer an alternative to conviction.
User experiences that align habit and intent
User reviews are deceptively practical. The traditional format is fast to write down however gentle to put in writing badly. A properly story avoids solution bias, describes the set off and the result, and is testable. The popularity standards do the heavy lifting. ChatGPT can write respectable reports quickly, but you want to be particular about the persona, the adventure degree, and the authentic-international triggers.
Give the variety a slender slice of context. Describe the persona in two traces, paste a snippet of the prevailing UI circulate, and be aware the established frustration. Ask for three story versions that reflect numerous constraints. You may possibly say, make one adaptation that preserves 0 again-give up changes, every other that accepts a one-week API adjustment, and a 3rd that assumes we can upload a brand new journey to the analytics schema. This attitude surfaces business-offs in engineering phrases.
Good attractiveness criteria learn like a contract between product and QA. They will have to be deterministic, no longer poetic. Include the poor paths and bounds. If a placing is simplest visual to admins, say so. If fee limiting kicks in after five attempts in fifteen mins, lock it down. ChatGPT is first rate at enumerating side situations when you set the boundaries and embody the business regulation that subject.
When you are integrating with regulated records or complex pricing good judgment, instant the mannequin to provide a policy map. For illustration, ask it to list each and every blend of plan style, role, and function flag that could alter the person sense. You can then prune anything else inappropriate and turn the the rest into reputation standards.
A quick anecdote approximately speed and rework
On a workforce development a workflow automation function for gross sales ops, our first inner spec ballooned to eight pages with diagrams and a dozen user reviews. Two senior engineers examine it and have been left with three entirely unique interpretations of the brink habits. We rewrote with ChatGPT as a drafting companion. The on the spot incorporated an anonymized CRM schema segment, a transcript from a customer name, and a screenshot of the contemporary rule builder UI.
We requested for a spec that centered on 3 conversion movements, integrated decide-in conduct via default, and assumed no transformations to the webhook structure. The brand produced a two-page draft. We then requested it to argue towards the inspiration from the standpoint of a performance engineer tracking queue intensity. It flagged a backlog chance if we allowed cascading re-reviews. That made it into the non-goals and transformed one recognition criterion. Result: the following engineering overview took 25 minutes, now not 90, and we minimize a dash of transform.
Turning discovery notes into a specific thing the team can use
Discovery dumps are in which clarity goes to die. You have Figma links, name transcripts, a spreadsheet with survey responses, and a 1/2-baked chart from analytics. ChatGPT is high-quality at compressing this into something you can still proportion in planning without diluting the sign.
Paste a small sample of each enter, not the overall corpus. Summaries decay whilst the form faces too much repetitive textual content. Ask it to extract consumer intents and objections with verbatim rates, then cluster them. Follow with a request for a single-page quick focused on the desirable two intents and one counter-purpose you propose to disregard, with a line on why. The forget about line subjects as it facilitates stakeholders see that saying no became regarded, now not unintended.
Treat the output as scaffolding. Replace well-known phrasing with the language your clients the truth is use. Swap adjectives for numbers. ChatGPT is much more likely to put in writing “many users battle” than “14 of 23 surveyed failed to uncover the export button inside of 30 seconds.” The latter is what your group necessities to read.
Avoiding the basic failure modes
There are three predictable approaches this goes sideways. First, common tale templates that flatten context. Second, specifications bloated with redundant sections that no person keeps. Third, recognition standards that learn like advertising and marketing copy.
To counter the 1st, avert a snippets library of your product’s vocabulary. A few lines in your roles, feature flags, naming conventions, and your quirky error states will shop the brand grounded. Paste snippets into your activates. The distinction reveals up in the pronouns and the verbs.
On bloat, set complicated size limits and ask for omissions. A worthy trick is to invite for “simply the sections necessary for kickoff,” then record the ones you want. There is no prize for which include a “Dependencies” header that stays empty for the subsequent two years.
For recognition standards, push the version toward if-then language and finite values. Avoid “may want to” and “may want to.” Ask it to enumerate detrimental paths and parameter limits explicitly. If considered necessary, supply it a skeleton with a few truly values and request of entirety that preserves the development.
Acceptance criteria approaches that carry up under test
Here is a small record that has kept me greater bugs than any single tool. Use it to sanity money sort-generated criteria sooner than handing them off.
- Define visibility, triggers, and put up-condition country explicitly, with at least one destructive case for each and every. Include limits for lengths, counts, timeouts, and rates, however provisional, so engineers ask about them. Capture function and permission interactions in simple terms and map them for your truly roles, now not hypothetical ones. Specify what gets logged or instrumented for analytics, with tournament names and assets keys. Call out what will no longer come about, highly detrimental moves that stay as-is.
Run due to these five elements and you will trap 80 p.c. of the anomaly that produces production surprises. ChatGPT can pre-populate them. Your task is to exchange defaults and add the one or two constraints original for your manner.
Writing for varied stakeholders devoid of whiplash
PMs write for engineers, designers, QA, revenues, executives, and frequently valued clientele. Each audience has one-of-a-kind thresholds for element and a distinct type of patience. The mistake is to supply one artifact and ship it all over. The bigger sample is to keep a core spec plus tailor-made summaries.
ChatGPT is robust at tone transferring. Feed it the core spec and furnish a quick persona transient for the viewers. Ask it to produce a one-paragraph replace for the government channel that names the enterprise consequence, possibility, and date. For revenue, request a two-paragraph notice with earlier than-and-after screenshots and 3 strains of effect on objection handling. For QA, ask for a matrix of states and transitions, in text, no longer a picture, that suits how they imagine.
Avoid upbeat adjectives in those outputs. They minimize credibility with senior readers. Replace them with concrete references to the plan. The kind will preserve looking to upload positivity. Edit that out.
Using ChatGPT for part case discovery
One of the top cost uses is to generate a catalog of aspect situations at once. Given your constraints, ask for techniques the approach may just fail in a well mannered way. For example, for a bulk import function, instant the form along with your box limits, record length cap, and error coping with layout. Request aspect instances grouped with the aid of person mistake, network habit, and details structure. You gets instances like “UTF-8 BOM at commence of CSV,” “blended line endings,” or “header capitalization mismatch.” Compare the listing on your truly-world logs and prune.
For troublesome permission systems, ask the edition to enumerate eventualities the place inherited permissions and direct provides struggle. If you provide your position hierarchy and inheritance regulations, it is able to generate tale-sized instances in mins. Again, overview ruthlessly. The target is to capture blind spots, now not to just accept the list wholesale.
Measuring influence with no slipping into shallowness metrics
PMs are answerable for outcomes. If you undertake ChatGPT on your writing glide, you should always degree even if it reduces cycle time and defects. The most simple degree is evaluation loop remember. Track how many overview cycles a spec calls for to achieve engineering signal-off. If that drops with the aid of even a 3rd, you might be saving days consistent with undertaking.
Another indicator is the illness combination put up-unlock. Look on the ratio of habit misunderstandings versus implementation bugs. If popularity standards get clearer, you must always see a drop in false impression-associated issues. It will now not swap in a single day, however over two or 3 sprints the development indicates up.
You could also measure internal time saved. If you preserve calendar blocks for spec drafting, notice entire hours in line with artifact prior to and after. Most teams I actually have worked with see a 20 to 40 percentage discount in drafting time. The truly win, although, is less to come back-and-forth, that's more difficult to quantify but very visual for your Slack history and standups.
Privacy, compliance, and what not to paste
Be careful with delicate info. Do now not paste shopper names, PII, or proprietary algorithms into a third-get together instrument. Redact or update with placeholders. If your service provider has an accepted atmosphere for ChatGPT use, stick to it. If Technology no longer, write prompts that embrace schemas and systems rather than raw info, and mask identifiers persistently. A dependable development is to stay a regional word with the unredacted context and change in placeholders sooner than sending the advised.
Also believe criminal commitments. If your targeted visitor contracts preclude tips sharing, translating reinforce tickets into anonymized charges can still be hazardous if blended with other alerts. When in doubt, work with your security crew to establish a sanitization record.
A reasonable prompting trend you may reuse
The maximum risk-free effects come from based prompts along with your personal language, rough constraints, and a clear output shape. Here is a trend I use repeatedly with properly effects. Adjust the tokens to suit your product.
- Context: one paragraph hassle announcement for your words, plus two to 4 short documents elements or fees. Constraints: the main limits and non-objectives that rely for this alteration. Audience: who this is for and what they need to do with it. Output form: sections you want, with individual or sentence limits in keeping with area. Pushback: ask for two dangers or disagreements from a described attitude.
A skeleton appears like this in simple textual content:
Context: Sales managers is not going to bulk reassign leads across territories with no exporting and reimporting CSVs. This explanations a weekly backlog and 12 p.c. lead aging beyond SLA. Support tickets prove frustration with errors messaging on import screw ups.
Constraints: no ameliorations to territory suggestions engine this zone, CSV add pipeline need to continue to be, no heritage jobs longer than five minutes, audit log required for each one reassignment.
Audience: backend and frontend engineers for kickoff, QA for try out planning.
Output: a spec lower than 700 phrases with sections for Scope, Non-ambitions, User Impact, Acceptance Criteria (eight to twelve lines), Metrics, and Risks. Use our terms: Territory, Lead Owner, Audit Event. Do no longer encompass architecture diagrams.
Pushback: consist of two negative aspects from the viewpoint of the SRE on queue load and a mitigation in step with possibility.
Feed it two or three authentic mistakes messages and a screenshot description of the latest import conversation. The first draft you get is in general 80 percentage of the approach there. You conclude the final 20 percentage.
When to dodge ChatGPT and write by using hand
If you are exploring a delicate strategic shift, writing a memo that argues for a extensive bet, or facing a subtle industry-off with workers effects, write the first draft your self. Models generally tend to smooth the rims of laborious judgements. Your voice issues in those situations. Bring ChatGPT in later to tighten sentences or to create stakeholder-targeted summaries, not to supply the center argument.
Another state of affairs is whilst the area has authorized or protection implications. Medical dosing calculators, economic counsel engines, or anything that touches public security merits area proficient authorship, then based QA, then felony overview. The edition can lend a hand with formatting and guidelines new release, but it should always not be requested to produce authoritative recognition standards with no heavy supervision.
Maintaining a living spec library
The worst destiny for a spec library is to change into a graveyard of stale archives. If you would accelerate advent, you also want a renovation plan. I preserve a temporary “spec delta” segment at the accurate of every artifact where adjustments are logged with dates and a one-line intent. Once 1 / 4, we archive the rest untouched for 90 days or extra into a study-simplest folder and hold forward purely what is lively. ChatGPT is incredible here for generating diffs when you paste the vintage and new variants and ask for a replace log in three strains. That makes stakeholder updates more easy and allows engineers see what replaced with out rereading the complete record.
Bringing the staff along
If you undertake ChatGPT for specs and thoughts, make it visual to your crew. Share activates and drafts. Invite engineers to signify set off improvements and to contribute their own templates. When engineers see their vocabulary contemplated within the output, they have confidence it greater. Designers generally relish swifter iterations on UX replica and empty-country language. QA groups improvement while popularity criteria apply consistent styles. Adoption grows while each serve as sees an instantaneous benefit.
The different cultural transfer is to treat the model as a junior collaborator, not an oracle. Encourage war of words with its drafts. Ask teammates to focus on anything else that sounds normal or unsure. Reward edits that add specificity. Over time, your activates becomes greater compact, your styles Complete guide to chatgpt in Nigeria greater secure, and your team will spend much less time wrangling phrases and extra time constructing.
Final options from the field
Specs and consumer experiences are wherein method meets fact. The craft is in turning messy inputs into transparent directions with out suffocating creativity. ChatGPT is a robust instrument for that craft when you operate it with guardrails and taste. Anchor it along with your product’s fact, deliver it constraints that replicate your global, and ask it to argue with you prior to you commit.
The benefits compound. Drafts come collectively swifter. Reviews tighten. Edge cases get stuck in the past. And possibly most importantly, the crew spends much less electricity deciphering rationale and greater energy fixing precise complications for users. That is the element of the task.
If you go with one addiction in the beginning, make it this: write your drawback fact and constraints to your personal words, paste two true artifacts, and ask for a brief spec with challenging limits and pushback. Edit ruthlessly. Share early. You will really feel the big difference the first week and notice it within the work by the conclusion of the month.