Most people building a digital product spend six weeks on the product and about ninety seconds on the price. They land on £19 because it feels safe, or £97 because someone on YouTube said to charge £97, and then they never work out whether the number was the problem.
Pricing is a research job, not a vibes job — and it is one of the few business tasks AI is genuinely excellent at, because the raw material is public and scattered across a hundred pages nobody wants to read. Here is the process I actually use.
Why the Guessed Price Is Usually Wrong in Both Directions
Underpricing is the obvious failure: you sell forty copies at £9 and conclude there is no market, when in fact you built a £60 product and charged pocket-money for it. Overpricing does something worse — it produces silence, and silence tells you nothing. A product at the wrong high price and a product nobody wants look identical from outside.
The fix is not to find the "correct" price — there isn't one. It is to arrive at a defensible range, then test inside it. AI gets you there in an afternoon instead of a fortnight.
Step 1: Map the Real Market, Not the Aspirational One
Your competitor set is not the polished course everyone quotes. It is what your specific buyer would actually consider instead of you — including free options, a book, and doing nothing.
Use a model with live web access (ChatGPT with search, Claude with web search, or Perplexity) so you get current listings rather than remembered ones:
That last sentence matters. Without it you will get plausible, confidently-stated prices for products that do not exist. Open every link. Anything you cannot verify gets deleted from the sheet.
Step 2: Price the Outcome, Not the Contents
Buyers do not price your PDF count. They price the thing they get to avoid. Work out what the problem currently costs your buyer in money, time or risk, and your price becomes a comparison instead of a number in a vacuum.
A revision resource competes with a tutor at £30 an hour. A template pack competes with four hours of a freelancer's time. A checklist that prevents one expensive mistake is priced against the mistake. Write that sentence down — "this replaces X, which costs Y" — and if you cannot finish it, your problem is the product, not the price.
Step 3: Build Three Tiers Deliberately
A single price gives the buyer a yes/no decision. Three prices turn it into a which-one decision, which is a far easier conversation to win. The standard shape: a stripped entry option, a main offer you actually want people to buy, and a premium tier that mostly exists to make the middle one look sensible.
Treat the output as a first draft. AI is good at structure and bad at knowing what delivery actually costs you in hours.
Step 4: Sanity-Check Against Your Own Numbers
Run the arithmetic before you commit. At each candidate price, how many sales do you need per month to make this worth the hours you are putting in? If the answer is 300 units and your email list has 60 people on it, the price is too low regardless of what the market says.
Count what each sale costs you too: platform fees, payment processing, support emails, refunds. A £12 product with a fifteen-minute support conversation attached is not a £12 product.
Step 5: Test the Price Where People Can Actually See It
Pricing only becomes real on a page. Put the tiers live, drive a small amount of traffic, and watch which one people click before they abandon. That single behaviour tells you more than any amount of research — and it only works if the page loads fast, reads clearly on a phone, and puts the comparison in front of the buyer instead of burying it. If your sales page is the bottleneck, that is exactly the kind of thing I fix in my web design work.
Change one price at a time, give it at least a hundred visitors, and resist discounting the moment it goes quiet. Discounting is how you teach an audience to wait for a sale.
The Afternoon Version
Build the verified competitor table. Write the "this replaces X, which costs Y" sentence. Draft three tiers. Check the maths against your actual audience size. Publish and leave it alone for two weeks.
Four hours of work — and it replaces the commonest reason good digital products quietly fail: a number picked in ninety seconds and never questioned again.