TL;DR
Paste your raw initiative list, three strategic themes, and the items you dropped into the prompt below. The AI returns a one-page summary tiered as Committed / Exploration / Not on this roadmap with at most 5 committed items per quarter. The one section that builds trust is “Not on this roadmap”: the AI cannot invent it, so you have to supply the dropped items and reasons yourself. For a 30-line roadmap, any current frontier model handles this in one pass; use Claude Opus 4.7 or Gemini 3.1 Pro when the source doc runs into hundreds of pages.
The task
It is roadmap-summary week. You maintain the canonical roadmap, a Notion grid or Linear cycle plan with 30+ initiatives across the year, color-coded by team and tagged by quarter. Your VP wants a one-page summary for the board, your CEO wants the same thing minus the engineering language, and your design lead wants to know why their three exploration items got “demoted” to the back page. Everyone is going to read the summary as a contract. You want a one-pager that is honest about what is actually committed, what is genuinely being explored, and (the part that builds trust) what is not on the roadmap and why.
This maps directly onto the Now-Next-Later structure that has largely replaced fixed-date timeline roadmaps in 2026: committed items are your “Now,” planned items are “Next,” and exploration items are “Later.” The prompts below produce that tiering automatically and refuse to attach hard ship dates to anything past “Committed.”
Where AI helps, and where it does not
AI is strong at compressing 30 initiative lines into 3 thematic clusters, tiering items by commitment level (committed / planned / exploration), and producing audience-specific phrasing for board vs. all-company vs. team. It is also good at flagging the “everything looks committed” failure mode and forcing you to demote items until at most 5 per quarter are truly committed.
What AI cannot do: decide which initiatives are actually committed. That is a judgment call about staffing, dependencies, and political risk, so pass it in as input. The other thing AI cannot do is invent the “not on the roadmap” list. Those are the items you considered and dropped, and the model only knows what you tell it. That section is the trust signal of the entire document; do not skip it.
A specific failure mode: AI will quietly promote “exploration” items to “planned” if you ask for “a cleaner narrative.” Do not let it. Once an exploration item gets a ship date in a summary, the team owns it whether or not the work was scoped, and you find out at the next QBR.
What to feed the AI
- Initiative list with: name, theme, status (committed / planned / exploration), expected ship quarter, expected outcome in one phrase
- The 3 strategic themes for the year (so initiatives ladder up to themes, not the other way around)
- The audience: board / leadership / all-company / team (each gets different granularity and different language)
- Last roadmap summary (for continuity, since leadership reads them as a sequence, not as standalone)
- Items you considered and dropped, with the reason (this is what makes the “not on the roadmap” section credible)
- The top 2 risks that could force a re-baseline (key hire delay, vendor blocker, market shift)
- Capacity constraints you want surfaced honestly (headcount, on-call burden, holiday weeks)
- The one sentence leadership should remember after closing the doc, usually a theme-level claim about the year
Copy-ready prompt
Placeholders are written as [...]. Replace each bracket, including the brackets themselves, with your own content before sending.
Summarize this roadmap for [board / leadership / all-company / team].
Strategic themes for the year (top 3): [paste]
Initiatives (name, theme, status, ship quarter, expected outcome):
[paste list]
Items considered and dropped, with reason: [paste]
Top 2 re-baseline risks: [paste]
Capacity constraints to surface: [paste]
The one sentence leadership should remember: [sentence]
Last roadmap summary (for continuity): [paste]
Return in this structure:
1) Headline - one sentence on what the team is solving this year, theme-led, not initiative-led.
2) Per quarter (Q1-Q4) - 2-3 committed initiatives only, each with a one-line outcome and a "ship by [date]." Label this section "Committed."
3) Exploration - items being investigated, no ship dates, label clearly. "Spike in Q3" is fine; "ship in Q3" is not.
4) Not on this roadmap - important items we considered and dropped, with the reason. This section MUST exist; if it is empty, you have not finished the prioritization.
5) Risks - the top 2 things that would shift this and what we would do if they hit.
Rules:
- At most 5 committed items per quarter. If the list overflows, push items to Planned or Exploration.
- Do not upgrade Exploration to Planned for narrative neatness. If status is "Exploration," it stays Exploration.
- Outcomes are user/business outcomes, not feature names ("recover $80k ARR lost to churn," not "billing v2").
- One page maximum. If it is longer, you are listing, not summarizing.Shorter variant: board read-out paragraph only
Write the 120-word board version of this roadmap.
Themes: [top 3]. Committed initiatives per quarter (max 5 each): [list with outcomes]. The biggest re-baseline risk: [one].
Format: 1 sentence on themes. 1 sentence per quarter with the top committed item and outcome. 1 sentence on the risk. No exploration items in this version; the board does not read those.Which model to use
For a typical 30-to-60-initiative roadmap (a few pages of source text), this is a short-context task and any current frontier model produces a clean one-pager in one pass: GPT-5.5, Claude Opus 4.7, Claude Sonnet 4.6, or Gemini 3.1 Pro. As of June 2026, ChatGPT Plus ($20/mo) reads roughly 320 pages of in-app context, Claude Pro ($20/mo) and Gemini (Google AI Pro, $19.99/mo) both expose a 1M-token window, so any of them is more than enough for a roadmap doc.
Reach for the long-context models only when your source is huge, for example a full Notion export with every initiative’s spec inline, or a year of changelog you want re-summarized:
| Model | Context window | Best for here | Notes (as of June 2026) |
|---|---|---|---|
| Claude Opus 4.7 | 1M tokens | Tight, structured prose; high output ceiling (128K out) | Best for one-shot long-form drafts; strongest at holding tier labels |
| Gemini 3.1 Pro | 1M tokens | High-volume summarization, lowest cost | Cheapest of the three on the API ($2 / $12 per 1M in/out) |
| GPT-5.5 | 1M tokens | Ultra-long retrieval across 100s of pages | Strongest on long-context recall (MRCR v2) when the doc is enormous |
For the actual one-page summary the differences barely matter. They matter when the input is large. If you are already inside Linear, Jira, or Notion, their built-in AI (Linear’s agents, Atlassian Rovo, Notion AI) can read the source in place, but they tend to over-list. Paste into a chat model when you want the demotion discipline the prompt above enforces.
Sample output
A useful committed-per-quarter block: “Q3 — Committed: 1) Launch billing v2 (ship by Sep 15) — recovers ~$80K ARR currently lost to involuntary churn. 2) Self-serve seat management — unblocks the enterprise renewal cycle in Q4. 3) Search relevance v3 — moves W4 retention by 2-3 points based on the test we ran in July.”
A useful exploration line: “Exploration: AI-powered admin assistant. Spike planned in Q3, no commitment to ship. Decision gate: end of Q3, based on prototype usage with 5 design partners.”
A useful “not on this roadmap” entry: “Enterprise SSO — important and asked-for in two deals this quarter, but blocked on a security hire we are still recruiting. Revisited next planning cycle. If a deal larger than $200K ARR closes contingent on SSO, we will re-baseline.”
A useful headline: “This year is about turning the activation funnel into the growth engine: paid users in by Q2 (billing), retained by Q3 (search relevance), and expanded by Q4 (team plans).”
How to refine
- Force-tier the items: “Three tiers only: Committed / Planned / Exploration. At most 5 Committed per quarter. If the list overflows, push lower-conviction items to Planned. Do not upgrade Exploration to Planned to make the quarter look fuller.”
- Outcomes over feature names: “Replace every feature name in the summary with the user or business outcome. ‘Billing v2’ is invisible to leadership; ‘recover $80K ARR lost to churn’ is not.”
- Demand the not-on-roadmap section: “If the ‘not on this roadmap’ list is empty, the prioritization is unfinished. Re-prompt with the list of items we considered and dropped, including the political ones (Enterprise SSO, AI-everything, etc.).”
- Compress to themes: “If a quarter has 4+ committed items, group two of them under one theme sentence and let the items be sub-bullets. Leadership remembers themes, not initiative names.”
- Match the audience’s vocabulary: “Board: ARR, retention, gross margin. All-company: customer language, no internal codenames. Team: scope, owner, dependency. Same body, three vocabularies.”
Common mistakes
- Treating the roadmap as a commitment-by-default: leadership stops believing it after the second slip; tier honestly and the document keeps its weight
- No “not on this roadmap” section: leadership reads this as “you have unlimited capacity,” which means every new request lands on your team
- Quarter-by-quarter listing with no theme: leadership cannot remember what the year is about; themes are the anchor that survives the meeting
- Putting ship dates on exploration items: the team now owns those dates whether or not the work was actually scoped; explorations get “spike,” not “ship”
- Letting capacity stay invisible: if Q3 is 60% of staff because of holidays and on-call rotations, that belongs in the summary, not buried in a Notion sub-page
- Updating the roadmap silently: re-publish quarterly even if nothing changed; silence reads as drift, and the team starts inventing their own version
- More than 5 committed items per quarter: you are not committing, you are listing; at 7+, leadership reads everything as “maybe”
- Letting AI rewrite “exploration” to “planned”: happens when you ask for “tighter narrative”; protect the tier labels above narrative neatness
FAQ
- Should I include exact dates?: Only for committed items, and only at week granularity (“ship by Sep 15”). Exploration items get “spike in Q3,” never “ship in Q3.” Planned items get a quarter, not a week.
- How often should I update the summary?: Re-summarize when commitments change (a cut, a slip, a new add). Re-publish the full version each quarter even if nothing changed; silence reads as drift, and the team starts working from outdated copies.
- What about competitive items I cannot discuss in writing?: Use a generic theme label (“a category bet we are exploring, details in person”) and discuss in the meeting. Naming a sensitive item in a doc leaks; omitting it entirely looks like the roadmap is incomplete.
- The model keeps saying every quarter has 6-7 committed items, what changes?: Add: “Cap committed at 5 per quarter. If you have more, the lowest-conviction items get moved to Planned. Force this cap before producing output.” Then re-run.
- What if a key dependency is from another team?: Surface it in the risks section, not in the initiative itself. Initiatives whose ship date depends on a team you do not control are not committed; they are planned, at best.
- Should the summary use Now-Next-Later instead of quarters?: For an external or all-company audience, yes. Map Committed to “Now,” Planned to “Next,” Exploration to “Later,” and drop the week-level dates. For the board, keep the quarter labels; they expect the time horizon. See ProdPad’s Now-Next-Later origin write-up for why fixed dates erode trust.
Related
- AI roadmap planning
- AI feature prioritization
- Roadmap Planning Prompts: Quarterly Plans That Survive Contact
- AI Drafts Your Project Status Update
- AI Sprint Planning
Tags: #AI writing #Product #Workflow #Roadmap