AI vs Human Writer Cost Calculator

Punch in your monthly word target, pick a model, set what a human writer charges and see the two bills side by side. The number you'll get tells you the raw cost of producing draft text. The long-form section below tells you what you still have to pay for around it, because content is more than keystrokes and we'd rather be honest about that than sell you the cheap answer.

Explain like I'm 5 (what even is this calculator?)

You want some words written. A robot will do it for pennies but might get the facts wrong. A human will do it well but charges proper money. This tool shows what each one costs for the volume you're after, and where the crossover sits. The robot bill is real. The human bill is real. The bit nobody puts on a sticker is the editing time, and we talk about that further down.

Calculate

Enter your numbers, then press Calculate.

Prove it

Output tokens are estimated at 1.3 tokens per English word (the four-characters-per-token rule of thumb applied to the average five-character English word plus a trailing space). LLM cost per article is (input tokens ÷ 1M × input price) + (output tokens ÷ 1M × output price). Monthly cost is cost per article × articles per month, where articles per month is the monthly word target divided by article length. Human per-word cost is taken straight from the rate you entered, or computed from hourly rate ÷ words per hour. Pricing is the standard non-cached, non-batch list price published by each provider.

Useful? Save this calculator: press Ctrl + D to bookmark it.

What this calculator is actually doing

Both columns are linear: cost goes up in proportion to wordcount. The LLM column multiplies your monthly word target by the per-word output cost of the chosen model, then adds the prompt overhead for each article (the system prompt and brief you send in). The human column either multiplies by the per-word rate you entered, or works out the hours (words ÷ words-per-hour) and multiplies by the hourly rate. There is no fixed cost on either side: no platform fees, no minimums, no agency retainers. Real life has all of those, so treat the figure as the per-word cost-of-production floor, not the bottom line.

The bit the LLM bill does not include

If you treat draft text as "done", the LLM column wins on price for almost any volume. It's a hundred times cheaper per word than even a budget freelancer. But draft text is not done. Here is what an honest content operation pays for around the keystrokes:

  • Editorial review. Someone reads it, fact-checks claims, removes the AI tells (the over-confident summary sentences, the four-word bullet points that all start with the same verb, the made-up statistics), and rewrites the bits that sound like the model phoned it in. Budget 30 to 90 minutes per 1,000-word article.
  • Subject-matter accuracy. The model will confidently state things that are wrong. Tax thresholds, drug dosages, legal precedents, product features. If your topic has any regulatory or safety dimension, you need a human who actually knows it to read the draft.
  • Voice and brand. LLM prose has a default register (slightly American, slightly LinkedIn, slightly hollow). Pulling it back to your brand voice takes work. Some brands genuinely cannot publish LLM prose without it landing as off-brand.
  • SEO sanity. The model will pad. It will repeat itself. It will use the wrong heading hierarchy. It will invent links. A human or a properly tuned editorial pass fixes this before publish.
  • Originality and authority. Search engines and readers are getting better at detecting and discounting commodity AI content. The pieces that earn links and rankings are the ones with original reporting, expert interviews, or a clear point of view, none of which come out of a base model.

Add the editorial time to the LLM column at whatever your editor's rate is. On a 1,000-word article at one hour of editing at $60/hour, that is another $60 on top of a few cents of API spend. The maths still favours the LLM at scale, but by a much smaller margin than the raw bill suggests.

When LLMs make sense

The LLM column wins clearly when:

  • The topic is well-trodden and the model has read about it a thousand times (general explainers, definitions, summaries of public knowledge).
  • You need volume that no human team could keep up with (programmatic SEO pages, product description variants, internal documentation).
  • The output is a starting point for a human, not the finished thing (draft outlines, first-pass briefs, research notes).
  • You have a strong editorial process to catch the inaccuracies and the AI tells before publish.
  • The downside of being wrong on any individual page is low.

When humans make sense

The human column wins clearly when:

  • The piece needs original reporting, lived experience or expert interviews. The model has none of those.
  • Accuracy under scrutiny matters: regulated industries, medical, legal, financial advice, anything that might be cited by a journalist or a regulator.
  • The brand has a recognisable voice that readers come for, and that voice is part of the product.
  • You're publishing rarely enough that one bad piece dents trust for the next year.
  • Your audience is clued-up enough to spot LLM prose and treat it as a signal of laziness.

Honest caveats on the numbers

Token counts are approximate

The 1.3-tokens-per-word figure is the rule-of-thumb applied to English prose. Code, JSON, tables, non-Latin scripts and very short articles will drift further. Treat the LLM cost as a sensible estimate, not an exact figure.

Caching and batch pricing not modelled

Anthropic, OpenAI and Google all offer prompt caching, which is genuinely useful here: most content workflows use the same long system prompt for every article. Caching can knock 50 to 90 percent off repeated input tokens. If you have a stable system prompt and high volume, your real bill will be lower than this calculator says.

Fine-tuned and reasoning models

The pricing here is for the standard chat models. Fine-tuned variants cost more per token and have a one-off training fee. Reasoning models (o1-style) charge for their internal thinking tokens, which can blow past output tokens by an order of magnitude on a hard task.

Quality is not in the spreadsheet

Two pieces with the same word count and the same publish date can be worth wildly different amounts to the business. The cheapest column does not always win; sometimes the expensive piece is the only one that earns a link, ranks for the high-intent keyword, or convinces a buyer.

Related calculators

Other tools for costing AI and content work.

Frequently asked questions

Is the LLM number really comparable to hiring a writer?

Not exactly. The LLM cost is the raw API bill for generating draft text. A real publishing operation also pays for the editor who fact-checks, restructures, fixes the voice and removes the AI tells. A human writer's price already bakes those things in. To compare like with like, add an editorial cost to the LLM column equal to the time you'll spend cleaning each draft. For most teams that's between 30 minutes and two hours per article.

Why does the per-word LLM cost look so low?

Because raw token output is genuinely cheap. A 1,000-word article on a mid-tier model is a few cents of API spend. The expensive bit of content is the thinking, the research, the editorial judgement and the legal review, not the keystrokes. The calculator shows the keystrokes cost honestly, then the long-form section talks about everything you still have to pay for around it.

What human writer rate should I plug in?

For UK and US freelance content writers in 2026, $0.10 to $0.25 per word is the working range. Specialist writers (B2B SaaS, finance, medical) charge $0.30 to $1.00 per word. Cheap content-mill rates ($0.02 to $0.05) buy what they buy. If you pay hourly, $40 to $80/hour at 400 to 600 finished words per hour is the realistic band for a competent writer.

Does this include caching, fine-tuning or batch discounts?

No. The figures use standard, non-cached, non-batch list prices. Prompt caching can knock 50 to 90 percent off repeated input tokens, which matters if your system prompt is long and shared across every article. Batch APIs cut another 50 percent in exchange for slower turnaround. If you are running this at scale with a stable system prompt, expect your real bill to be lower than the calculator suggests.

When does a human writer make sense over an LLM?

When the topic needs original reporting, expert interviews, lived experience, legal accuracy, or a recognisable voice. When the brand is small enough that one bad piece dents trust. When the audience can spot AI prose at fifty paces. When you need an opinion that holds up under scrutiny. The LLM column wins on volume and speed; the human column wins on credibility and original thought. Most serious publishers use both and are honest about which is which.

Why is the breakeven sometimes "human is never cheaper"?

Because both costs scale linearly with word volume in this calculator: there is no fixed-cost base for either side, just a per-word rate. If the LLM per-word rate is below the human per-word rate, the human side will always cost more, no matter the volume. The calculator surfaces that honestly rather than inventing a fake breakeven number.

Does this calculator send my data anywhere?

No. Everything runs in your browser. Your numbers never leave your device.

Last updated 29 April 2026 by Andrew Laws.