TaskbyTask.Practical AI guides for work
ChatGPT or Claude · Practical walkthrough

Choose ChatGPT or Claude for your business

Buying AI for work means choosing an account, a set of tools and a bill. A polished answer to one prompt is not a purchasing strategy.

Task by Task13 min read + practice
Two equal work trays hold matching briefs while a pencil marks a shared evaluation sheet.

What you'll make

a shortlist and a costed seat mix for your own work, with the features and permissions still to confirm before you buy.

What you'll need

your three main jobs, the AI providers your organisation already allows, the number of users and a rough monthly budget. Allow about 30 minutes. Use public or fictional material while you practise.

Example result.

Job Shortlist first Still to confirm
Product notes → editable deck with an original illustration ChatGPT Image use rights; deck export format
Policy questions answered from the handbook Either Which plan connects to the handbook folder
Ten seats, eight light and two heavy users Either Local GBP checkout price; overage policy

1. Write down the three jobs

ChatGPT (OpenAI) and Claude (Anthropic) both write, research, read files and help with technical work, so choose on capabilities, account terms, workflow and cost — not reputation. For each recurring job, name the input, the finished output, where it lives and who checks it. “Help with marketing” is too broad. “Turn product notes into an editable presentation and generate an original illustration” gives you something to assess. Paste your notes into a work chat:

Help me prepare an AI-tool shortlist for these three jobs:
[jobs]

Our required account and data controls are:
[controls]

Our checked plan information is:
[plans]

Return a table of requirements, evidence I have given you, gaps to verify,
and total subscription commitment for [users] users.
Separate essential requirements from preferences.
Do not invent prices, integrations, permissions or product test results.
Do not choose a winner where an essential requirement is unverified.

[jobs] is your three job notes. [controls] is the providers and data you may use. [plans] is the plan names, seat prices and billing period copied from the providers’ pages. [users] is the head count.

Why this works: the last line stops a fluent answer from filling an evidence gap with a recommendation.

What you should see: a table with a “gaps to verify” column that isn’t empty. If every cell is filled in, something has been assumed.

2. Shortlist by the capability that changes the decision

Most jobs don’t separate the two. These do:

If the job needs… Shortlist
Original illustrations or photo edits ChatGPT, which generates and edits images. Claude analyses images and builds HTML/SVG diagrams but doesn’t generate pictures.
A finished report, deck or file Either: ChatGPT’s Work switch or Claude’s file creation. Check your audience can open the format.
Knowledge in your work systems Either: check the exact plugin or connector, and whether it reads, writes or both.
Work that lives in Microsoft 365 or Google Workspace Add your Copilot or Gemini option.

Availability depends on the plan, region, rollout and administrator. If your organisation already runs one provider and it meets the essentials, start there.

Why this works: it rules setups out on the essentials before anyone argues about which one “writes better”.

What you should see: each of your three jobs marked ChatGPT, Claude, either or “neither, use what we have”.

Claude’s Docs, Slides and Design templates are available on all plans. They need Cloud code execution and file creation enabled in the capability settings. Enterprise turns them on by default on 15 October 2026; an owner can opt in earlier.

3. Cost the seat mix

Both business plans publish the same USD seat prices, in two seat types you can mix. The pound figures below are budget estimates using the HMRC monthly rate (October 2026: £1 = $1.3456), not UK checkout prices.

Seat (ChatGPT Business or Claude Team) Monthly billing Annual billing, per month
Standard $25 (approx. £18.58) $20 (approx. £14.86)
Premium $125 (approx. £92.90) $100 (approx. £74.32)

Both need at least two seats; Claude Team runs from 2 to 150. Premium buys more capacity, not less checking. Enterprise is quoted separately, and Claude prices it as a seat fee plus usage at API rates, so a seat-only comparison misses part of that bill.

Use the price shown for your billing address, since both providers handle currency and tax by region. If 20% VAT is added on top, £14.86 becomes about £17.84.

Team subscription calculator

Start with a GBP budget estimate, or switch to the USD reference. The seat prices are the same for ChatGPT Business and Claude Team; their usage allowances are not.

Approx. £267.54/month equivalent · £3,210.46 annual commitment, before tax for eight Standard and two Premium seats on annual billing.

GBP estimates use £1 = $1.3456 (HMRC, October 2026). Totals use unrounded rates, then round to pennies. Excludes card fees, API charges and extra usage.

Then ask about extra usage, separately billed API work, paid connected services, training, and how leavers affect an annual commitment. The subscription is the most visible line in the budget. It is rarely the only one.

Why this works: pricing a mix forces the question of who needs heavy capacity.

What you should see: a monthly and annual figure, before tax, to set beside a real quote.

4. Confirm the team controls

Compare managed organisation plans, not personal subscriptions: a work email address doesn’t make a personal account a company workspace. OpenAI says it doesn’t train on ChatGPT Business data, and Claude lists no training by default for Team. Neither means zero retention.

Ask your administrator about:

  1. People: who gets a seat, who administers it and how leavers lose access.
  2. Data: what may be shared, retention, residency and contract terms.
  3. Connections: which systems can be read or changed, and who authorises each one.
  4. Oversight: whether the audit, identity and spend controls you need are in this plan, or only in Enterprise.

Why this works: these questions block a purchase late, after everyone has chosen a favourite.

What you should see: a yes, no or “Enterprise only” against each line.

Check the result

  • Every price matches a figure you copied from the provider’s page this month.
  • Every “yes” has a source: a help page, your admin console or a test you ran.
  • The total converts unrounded dollars once, at the end.
  • No job has a winner while an essential is still a gap.

The usual failure: the worksheet quietly fills a gap. It credits a provider with a feature it doesn’t have, or treats a converted pound figure as a confirmed checkout price. Spot it by asking where each “yes” came from.

Remove every capability and price from the worksheet that I did not give you.
Mark each one as unverified and state the specific evidence needed.
Recalculate using only the seat counts and rates I confirmed.

Worked example: a ten-person team

Fictional example. A ten-person agency has eight people doing document work and two doing heavier analysis. It pasted this into the worksheet prompt:

Jobs: 1) product notes → editable client deck with one original illustration
      2) answer staff questions from the handbook
      3) monthly analysis of a 5,000-row sales export (two people, daily)
Controls: both providers allowed; no client data in personal accounts
Plans: Standard $20/seat/month annual, Premium $100/seat/month annual
Users: 10 (8 Standard, 2 Premium)

The first answer recommended Claude Team for everything, “including the original illustrations”, and listed the budget as “£267.54 a month, confirmed”.

  1. The failure. Claude doesn’t generate illustrations, and £267.54 is a conversion, not a quote. The repair prompt moved both into “unverified”: job 1 now shortlists ChatGPT, and the price needs a local checkout figure.
  2. The arithmetic. 8 × $20 + 2 × $100 = $360 a month, or $4,320 a year: approx. £3,210.46 before tax. All ten on Standard would be $2,400, approx. £1,783.59. The two heavier seats add approx. £1,426.87 a year.
  3. The decision. Measure whether those two people actually hit a Standard limit on representative work before paying for Premium. The team should not buy ten Premium seats because two people hit a limit.

Fictional annual budget estimate in GBP before tax: ten Standard seats cost approximately £1,783.59; eight Standard plus two Premium cost approximately £3,210.46, a difference of £1,426.87. Converted at HMRC’s October 2026 rate of £1 to $1.3456.

Task kit for this job: Download (ZIP) · Reusable instructions

Go deeper

If two of you also need coding

A chat that explains code is different from a coding agent that edits a project and runs commands. For that, compare OpenAI’s Codex with Claude Code on a small existing project with clear acceptance criteria: correct results, readable changes, meaningful tests and a clean rollback. Settle four questions first:

  • Where does it run? Local or cloud decides which files and credentials it can reach.
  • Can it show the changes? You need a readable difference between old and new files, not a confident completion message.
  • What may it do? A file edit, a test, a Git push and a production deployment have different consequences. Keep deployment a separate, explicit step.
  • Which allowance pays? Long agent tasks use far more capacity than short chats.

Start with the Codex beginner’s guide.

Run a representative trial without inventing a winner

Once the essentials fit, try three jobs in both tools: a source-grounded report, a file task with a known answer and one connected workflow you need. Use the same inputs and acceptance criteria. Record the plan, settings, limits reached and how long checking took. Score factual errors separately from style, and repeat one awkward case. Treat the result as evidence about your jobs, not a universal ranking. The comparison exercise gives you a structure.

Keep the choice current

Keep a short decision record: plan, essential features, total commitment, missing evidence and a review trigger. Recheck when a feature you need ships, a price changes, your controls change, users keep hitting limits, or the HMRC rate moves enough to matter to your budget. A second subscription needs a specific job, such as original images or a connection the first tool lacks; don’t buy it for everyone when two people need it.

Sources

Optional analytics

With your permission, Google Analytics uses cookies to measure visits and which guides people read. It stays off until you accept. Use Analytics choices in the footer to change your choice. Rejecting after accepting refreshes this page. Privacy details.