I pay for AI tools every single month, and I’ve written before about the ones that have actually earned a place in my weekly routine : ChatGPT Plus, Claude Pro, and a couple of others besides. When you add the VAT on top, the bills stack up faster than you might expect. I did the maths recently and realised I was spending well over £40 a month across multiple subscriptions. That is a decent chunk of change for a working dad in Hampshire trying to be sensible about where the money goes. I use these tools daily, for the blog, for work, for brainstorming, for drafting things I would otherwise spend twice as long on. So they earn their keep. But the question I had never properly asked myself was: am I actually using them efficiently?
That question is what led me to build the AI Prompt and Token Efficiency Checklist over on the tools page . It was originally something I put together to help other people. Then I ran it against my own habits. I will be honest, some of what it flagged was genuinely surprising. Not catastrophic, but the kind of thing where you sit back and think, “Right, I have been doing that wrong for over a year.”
This article is what I found, why it matters, and how to fix it. Whether you are paying for one AI subscription or three, this is worth ten minutes of your time.
What Is a Token and Why Should You Care?
Most people paying for ChatGPT Plus or Claude Pro have no idea what a token is, and that is completely understandable. The services are sold as flat-rate subscriptions. You pay your monthly fee, you use the tool, job done. But tokens are what actually govern how much you can do before you hit a usage cap or start getting slower responses.
A token is roughly four characters of English text. Every word you type in a prompt costs input tokens. Every word the AI sends back costs output tokens. If your prompt is 100 tokens long and the AI writes a 200-token response, you have used 300 tokens total. The critical thing to understand is that output tokens cost significantly more than input tokens, typically three to five times more in commercial pricing terms. That has direct implications for how you ask questions.
On a flat-rate subscription, you do not see a bill for tokens. But they still govern your rate limits, your usage caps, and ultimately the quality of your experience. When you burn through tokens inefficiently, you hit limits faster, get throttled sooner, and end up with less value for the same monthly spend.
What the Checklist Actually Flagged in My Own Usage
Running the AI Prompt and Token Efficiency Checklist against my own habits was a bit like finding an old receipt for something you overpaid for. Not painful, just slightly annoying in retrospect.
The first thing it flagged was polite filler. I have been typing “please” and “thank you” into prompts for as long as I have been using these tools. It feels natural. You would not bark orders at a colleague. But it turns out this is genuinely wasteful. Research shows that non-polite prompts generate around 14 fewer tokens per request than polite ones. That sounds trivial until you think about how many prompts you send in a week. The AI is not offended if you skip the pleasantries. It does not need them. Dropping “please could you kindly write me a short summary if possible, thank you” down to “summarise this in three sentences” can cut the token cost by more than half and usually gets a better result.
The second flag was vague opening prompts. I caught myself doing this constantly, particularly when I was tired or rushing. Something like “summarise this article” gives the model almost nothing to work with. Too long? Too short? Written for a technical audience or a general one? Focused on which part? I would then spend the next three or four follow-up messages trying to steer the output into something useful. Each of those corrections costs tokens. Each one burns usage. I was doing this repeatedly when a single well-framed prompt would have got me there first time.
The third issue the checklist surfaced was the long conversation trap. This one surprised me the most. I had assumed that continuing a conversation was neutral in terms of efficiency, but it is not. Every message you send in a long thread carries the entire history of that conversation as context. By the time you are deep into a long session, even a three-word reply from you is dragging an enormous amount of prior text along with it. There is documented research showing that in extended Claude sessions, token costs for simple replies can balloon dramatically the further you go. The fix is simple. Start a new conversation for a new task. Do not just keep threading everything into one endless chat.
How to Prompt Better Without Becoming a Developer
None of this requires any technical knowledge. The checklist is designed for normal people, not engineers. But there are a few practical habits that make a real difference.
Be specific about format and length from the start. If you want three bullet points, say so. If you want a 200-word summary, specify it. This directly controls output tokens, which as we established are the expensive part. “Write me a blog intro” is a vague instruction. “Write a 100-word opening paragraph for a blog post aimed at UK parents who are new to AI tools, casual tone, no jargon” is a clear one. The second version will almost always produce something usable on the first attempt.
Give the AI a role or context where it helps. Not in a gimmicky way, but a single sentence of framing can dramatically improve relevance. “You are helping me draft a newsletter for a tech blog audience” costs almost nothing in tokens and saves multiple rounds of correction.
Match the model to the task. Not every job needs the most powerful AI available. For quick rewrites, simple summaries, and basic drafting, some of the free tools do the job well and burn through your paid limits more slowly. I have started defaulting to the faster, lighter options for routine tasks and reserving the heavier models for things that genuinely need them.
What Are You Actually Paying For? A Quick Subscription Comparison
To give this context, here is a quick snapshot of what the main AI subscriptions cost UK users right now.
| Service | Base Price | UK Cost After VAT | Notes |
|---|---|---|---|
| ChatGPT Plus | $20/month | ~£19.20/month | Fixed GBP price at checkout |
| Claude Pro | $20/month | ~£19.50–£21.00/month | Billed in USD, varies by exchange rate |
| Google AI Pro | $19.99/month | Not confirmed in GBP | Verify at checkout |
| Microsoft Copilot Pro | $19.99/month | Bundled with M365 Personal | Includes Office apps and OneDrive |
If you are running two or three of these simultaneously, you are looking at £40 to £60 a month after VAT. That is a real commitment, and it makes the question of efficiency much more important than most people realise. A quick tip if you are paying for Claude Pro: using a fee-free card like Monzo or Revolut saves you roughly £0.60 a month on currency conversion. That sounds small, but it adds up to nearly a full month’s subscription over the course of a year.
Hype Cycle Check
LIKELY TO LAST: Better prompting habits as a genuine skill. As AI tools become more embedded in everyday work and family life, the ability to get consistent, useful results efficiently is a durable advantage. This is not a passing trend.
WATCH CLOSELY: Usage-based pricing models becoming more visible to consumers. Right now, flat-rate subscribers have no dashboard showing their token consumption. If AI providers start surfacing this data, it will change how people interact with these tools significantly, and potentially introduce per-use billing for heavy users.
VAPOURWARE RISK: “Prompt engineering” as a formal profession. There was a period where this was seriously discussed as a major job category. The reality is that the models are getting better at interpreting natural language, which gradually reduces the premium on elaborate prompt crafting. The skill still matters, but probably not as a standalone career.
What This Means for CES 2027
AI efficiency is already one of the undercurrents building towards CES 2027. What I expect to see on the show floor is a shift away from raw capability announcements (more models, bigger context windows, faster outputs) and towards interface-level tools that help users get better results more reliably. Think prompt assistants built into consumer products, smarter conversation management within AI apps, and possibly the first consumer-facing token dashboards from the major providers. The conversation is moving from “look what AI can do” to “look how efficiently it does it.” That is a maturation worth watching.
What to Watch
Transparent usage dashboards for flat-rate subscribers. Right now you are flying blind. If OpenAI or Anthropic introduce token visibility for Plus and Pro users, it will fundamentally change how people prompt.
Lighter model quality closing the gap. The smaller, faster models are improving rapidly. As that gap narrows, the case for defaulting to cheaper tiers for routine tasks becomes much stronger.
Household AI budgeting tools. As families run multiple subscriptions across different devices, someone is going to build a clean tool for tracking and optimising AI spend. It does not exist properly yet.
Annual billing options. Claude Pro already offers around $17 a month on annual billing. If ChatGPT Plus follows suit, which it does not currently support, the maths changes considerably for committed users.
If you want to run these checks against your own habits, the AI Prompt and Token Efficiency Checklist is free to use at techdadslife.com/tools/ai-prompt-token-checklist . It takes about five minutes and will almost certainly flag something you had not thought about.
For more practical AI coverage like this, plus tech that actually works for families rather than just sounds impressive, sign up for the Tech Dads Life newsletter at techdadslife.beehiiv.com . No spam, no filler, just the useful stuff.
