Research and office delivery
Start here for source reading, synthesis, reports, presentations, and spreadsheets.
- Research material
- Write a report
- Build a table
- Prepare a brief
Finish one useful work deliverable with AI.
You do not need to become a programmer first. Bring a real task, give the AI clear facts and boundaries, then review the result like a responsible colleague.
What should I use right now?
Choose by the final deliverable, not by which product sounds more advanced.
Build, automate, and publish
Use Codex when the output is a website, code change, repeatable file operation, integration, or deployment.
- Build a page
- Process files
- Connect tools
- Deploy a site
Research first, publish second
Use Work to establish facts and approved copy. Hand the confirmed material to Codex to build and ship it.
- Confirm evidence
- Approve content
- Build the output
- Check the release
A safe combined path
Research output is not automatically approved publishing input. Keep a human checkpoint between the two.
- 1WorkCollect sources and draft the content.
- 2YouCorrect facts, remove private data, and approve the message.
- 3CodexTurn approved material into files, a site, or an automation.
- 4YouPreview, test, and explicitly approve publishing.
Three tasks you can finish today
Replace the bracketed fields with your facts. If a fact is missing, ask the AI to mark it unknown instead of guessing.
Marketing plan
Turn a business objective and a few known constraints into a reviewable plan, not a pile of slogans.
- Tool
- Start in Work
- First draft
- 30–45 min
Prepare before you start
- Objective and success metric
- Target audience and region
- Budget range and available channels
- Campaign dates and internal constraints
- Existing research, product facts, and approved claims
Copyable prompt
You are helping me draft a marketing plan. Use only the facts I provide and clearly label every missing item as “unknown — confirm with owner.” Do not invent market size, customer quotes, conversion rates, competitor claims, or sources.
Context
- Product/service: [name and one-sentence description]
- Business objective: [objective]
- Target audience and region: [audience]
- Budget range: [budget]
- Available channels: [channels]
- Time range: [dates]
- Confirmed facts and source links: [facts / links]
- Constraints: [legal, brand, staffing, approvals]
Please produce:
1. A one-page executive summary.
2. Audience assumptions separated into confirmed facts and hypotheses.
3. A channel plan with purpose, message, cadence, owner, budget, and success metric.
4. A week-by-week schedule.
5. Risks, dependencies, and open questions.
6. A fact-check table listing every number and claim, its source, and verification status.
End with the five questions I should answer before this plan is approved.Expected deliverables
- One-page executive summary
- Channel and budget table
- Weekly schedule
- Claim and source checklist
Human review checklist
- Do the objective and metrics match?
- Can every number and claim be traced to a source?
- Are owners, dates, and approval gates explicit?
- Have hypotheses been kept separate from known customer facts?
Privacy boundary
Remove customer names, private contact lists, unpublished pricing, credentials, and contract details. Use placeholders or approved aggregates.
Common failure
Asking for a ‘creative viral campaign’ before supplying an objective, audience, evidence, budget, or approval constraints.
Product report
Convert scattered notes and evidence into a decision document whose facts, risks, and unknowns are visible.
- Tool
- Start in Work
- First draft
- 35–60 min
Prepare before you start
- Report audience and decision to support
- Product brief or approved description
- Research notes, interviews, tickets, and metrics
- Date range and version being assessed
- Known limitations and sensitive fields to remove
Copyable prompt
Help me prepare a product report for [audience] to support this decision: [decision]. Analyze only the supplied material. Do not create user quotes, usage data, feature status, benchmarks, or citations. Mark conflicts and missing evidence explicitly.
Materials
- Product and version: [product / version]
- Reporting period: [dates]
- Product brief: [facts]
- User problems or research notes: [notes]
- Feature evidence and metrics: [data with source]
- Risks, incidents, or constraints: [facts]
- Source list: [links / file names]
Build the report with:
1. Executive summary and decision needed.
2. User problems, separating observed evidence from interpretation.
3. Feature analysis: intended outcome, current evidence, limitation, and confidence.
4. Risks, dependencies, and unresolved questions.
5. Recommendation with alternatives and trade-offs.
6. An appendix mapping each important statement to its source.
Before the report, list any material that is too weak or ambiguous to support a conclusion.Expected deliverables
- Decision-oriented summary
- User problem and feature analysis
- Risk and alternatives section
- Statement-to-source appendix
Human review checklist
- Does the report answer one defined decision?
- Are observation and interpretation visibly separated?
- Are versions and reporting dates correct?
- Can a reviewer find the source behind each major conclusion?
Privacy boundary
Redact names, email addresses, raw interview recordings, internal incident details, unreleased roadmaps, and account-level usage records unless sharing is approved.
Common failure
Mixing multiple product versions or treating a repeated opinion as measured user behavior.
Competitor comparison
Compare products on one consistent set of dimensions and make uncertainty part of the result.
- Tool
- Work, then Codex if publishing
- First draft
- 45–75 min
Prepare before you start
- Decision the comparison must support
- Products and exact plans or versions
- One shared set of comparison dimensions
- Dated primary source links
- Your own requirements and non-negotiables
Copyable prompt
Create a competitor comparison that supports this decision: [decision]. Compare these products and exact plans/versions: [list]. Use the same dimensions for every product. Use only the evidence I provide; do not infer absent features or fill gaps from memory. Mark unavailable or conflicting evidence as “unknown.”
Our requirements
- Must-have outcomes: [requirements]
- Budget and region: [constraints]
- Comparison dimensions: [for example price, core workflow, integrations, privacy, support, implementation effort]
- Evidence with access dates: [links / excerpts / files]
Produce:
1. A comparison table with evidence date and source for each material claim.
2. An unknowns and conflicts table.
3. Strengths and limitations for each product, written neutrally.
4. A weighted recommendation tied to our requirements, including the weight rationale.
5. Situations where a different option would be better.
6. A verification checklist for pricing, terms, and feature availability before purchase.
Do not use review counts, customer sentiment, market share, or performance claims unless those figures and their sources are supplied.Expected deliverables
- Comparable evidence table
- Unknowns and conflicts list
- Requirement-weighted recommendation
- Pre-purchase verification list
Human review checklist
- Are all products compared on identical dimensions?
- Are plan, region, currency, date, and version aligned?
- Do recommendation weights reflect our needs rather than popularity?
- Are unknown items visible instead of silently scored?
Privacy boundary
Do not paste confidential vendor quotes, negotiated discounts, account identifiers, private procurement notes, or credentials into an unapproved tool.
Common failure
Comparing a free plan with an enterprise plan, or turning missing evidence into a zero score without saying so.
The rule that matters most
AI can accelerate a draft. It cannot accept business responsibility for you. Verify numbers and sources, keep confidential material out, and approve every external action yourself.