AI Guides › Playbooks
By Nigel Guy · 6 min read
Most people who say "I can't afford AI" have never added up what it would cost. They have a feeling about headline numbers, a vague memory of a £200-a-month plan, and a ready-made reason to stand still. That feels like prudence while it quietly stops you testing anything.
On 30 July 2026 OpenAI gave that feeling a fresh jolt. It cut the API price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%, roughly three weeks after launch. That is real news, but the useful move is not to be impressed by it. It is to price your own use and see what is left standing.
The rule: Put a pound figure on your actual use before you say the word "afford", then name whatever is still blocking you.
Verified against OpenAI's own announcement of 30 July 2026:
Two things to hold loosely. First, "up to 80%" applies to one model, the cheap one, not to everything. Second, this moves fast: the current OpenAI API pricing page already lists a newer GPT-6 family, with Luna there at $0.10 input and $0.50 output per million tokens (about 8p and 38p at time of writing). Treat every figure here as a snapshot and check the live page before you budget.
A ledger has three columns and takes about twenty minutes. It separates money from everything else.
| Column | What you write | Why it matters |
|---|---|---|
| 1. Task | One repeated job you would hand to AI | Cost depends on the job, not on "AI" in general |
| 2. Monthly cost | A pound figure from the steps below | Replaces the feeling with a number |
| 3. Real blocker | Money, time, skill, trust, data or permission | Only one is allowed per row |
Pick jobs you do weekly or more: replying to enquiries, summarising meeting notes, tidying product descriptions. Not "transform my business".
Most individuals do not need the API at all. At time of writing OpenAI lists ChatGPT Free at $0 and Go at $8 a month (about £6), Plus at $20 (about £15) and Pro from $100 (about £75). US dollar prices were what the page showed to me, so check the £ price at checkout, including VAT. Free includes unlimited text chats with Luna subject to abuse guardrails, with limits on uploads, images and voice. Go is described as possibly including ads.
If a script or tool will call a model automatically, estimate tokens. A token is a small chunk of text, very roughly three-quarters of a word. Multiply your monthly input tokens and output tokens by the per-million price, and divide by a million. Look at "cached input" prices too, which are lower for repeated context.
For each row, finish this sentence with exactly one word from column 3: "If it were free tomorrow, the thing stopping me would be ____." If the answer is anything other than money, the price cut changes nothing for you, and that is useful to know.
For any row where money is not the blocker, take the cheapest route and run one real task this week. Set a pass mark in advance, such as "usable after two minutes of editing".
Priya runs a small bookkeeping practice in Leeds. Her three tasks are: drafting replies to client emails, summarising call notes, and turning receipts into tidy lists.
For the email task she imagines 200 replies a month, each roughly 1,500 tokens of input (the client email plus her house style) and 500 tokens of output. That is 300,000 input tokens and 100,000 output tokens. At the Luna API prices above, input costs about $0.06 and output about $0.12, so roughly $0.18 in total, about £0.14 a month at time of writing. Her honest figure for that row is pence.
Her ledger then reads: email replies, pence per month, real blocker is trust (client confidentiality). Call notes: £0 on the free route, blocker is skill (she does not know how to prompt it). Receipts: £0 to £15 depending on upload limits, blocker is time. Money did not appear once. She now knows what to work on.
You are a practical cost analyst helping a UK sole trader work out what AI would really cost them.
Goal: produce a three-column ledger (Task, Monthly cost in £, Real blocker) for the tasks below, so I can see whether money is genuinely the problem.
My inputs:
- Tasks I repeat: [TASK_LIST]
- How often I do each one per month: [FREQUENCY_PER_TASK]
- Typical length of the text going in and coming out: [APPROX_WORDS_IN_AND_OUT]
- Sensitive data involved (client details, health, financial): [SENSITIVE_DATA_YES_NO_PER_TASK]
- Tools I already pay for: [CURRENT_SUBSCRIPTIONS]
Steps:
1. If any input above is missing or vague, ask me for it before doing any sums. Do not guess.
2. Ask me to paste the current per-million-token prices and the current plan prices from the vendor's official pricing page. Do not rely on memory for prices.
3. For each task, estimate tokens (about 0.75 words per token), then calculate the monthly cost for (a) the free tier, (b) a flat-fee plan and (c) the API. Show the arithmetic.
4. Convert any dollar figure to pounds, label it "about £X, check at checkout" and say which exchange rate you used.
5. For each task, pick exactly one real blocker from: money, time, skill, trust, data, permission. Explain in one sentence.
6. Finish with the single cheapest test I could run this week, and a pass mark.
Format: a table, then a short paragraph of next steps. Plain British English, no hype.
Before answering, check that every number you used came from my inputs or the pasted prices, and that no row has more than one blocker.
Fill in the bracketed lines, then paste the official prices when it asks.