Your Listing Is Now a Source Document, Not a Keyword List.
If you take one mental model from everything I've written about AI discovery, make it this one, because it reorganises everything else: your listing is no longer a keyword list. It's a source document. It's the reference material an AI reads before it decides whether to vouch for your product to a customer who asked it a question.
Once that clicks, a lot of listing decisions that used to feel like guesswork suddenly have an obvious answer.
The old job vs the new job
The old job of a listing, in the keyword-matching era, was to contain the right words. A customer typed "insulated water bottle 32oz," the algorithm checked whether those words appeared in your listing, and surfaced you accordingly. Your listing's job was to be a well-stocked cupboard of terms. Crude, but that's genuinely how a lot of it worked, and it's why keyword-stuffing survived so long.
The new job is different in kind, not degree. When a shopper asks an AI assistant "which water bottle keeps drinks cold all day and fits in a car cup holder?", the AI reads your listing as evidence and asks: does this product actually satisfy that? Does the listing tell me it's insulated, how long it holds temperature, its diameter, whether it fits a cup holder? It's not checking whether you own the words. It's checking whether you've given it the facts it needs to recommend you with confidence.
That's the job of a source document. Not "contain the keywords" but "supply the evidence."
Why this is really just an old principle wearing new clothes
I've always argued that the keyword you choose is really the customer you invite — you're not optimising words, you're deciding which human, with which intent, you want to walk through your door, because that human's reaction is the lesson Amazon learns about you. The source-document era is that same principle made literal. The AI is standing in for the customer's question, and your listing either answers that question well enough to earn the recommendation or it doesn't.
So this isn't a new discipline to bolt on. It's the discipline I've been banging on about, finally with the crude keyword shortcut removed. Write to genuinely answer the question a real customer in your category is asking, and you're writing a good source document by default.
What "write it like a source document" means in practice
Concretely, it changes how you write, in a few clear ways.
Specifics beat adjectives. "Durable" is a marketing word; an AI can't do much with it. "Made from tear-resistant 600D nylon, tested to a 50kg load" is evidence — it lets the AI confidently answer "is this hard-wearing?" with a yes and a reason. Every vague claim you can replace with a specific, checkable fact makes you a better source.
Completeness is a feature. A gap in your listing is a question the AI can't answer about you — and a question it can't answer is a recommendation you don't get. If your category's buyers care about dimensions, materials, compatibility, care instructions, then a listing missing those isn't just thin, it's unquotable on those points. Fill the gaps that matter to real buying decisions.
Answer the actual questions. The best source material anticipates the questions. And you don't have to guess what they are, because Amazon hands them to you: look at the customer Q&A section Amazon posts on most listings — yours and your competitors' — alongside the review complaints and the "will this work for X?" queries. Those posted questions are a direct feed of what real buyers in your category are unsure about. One caveat worth its weight in gold: the answers in that Q&A section are sourced from other customers, and they are not always correct. So mine the questions for what people want to know, but verify the answers yourself before you build listing copy on them — because if a customer's wrong answer is sitting on your listing, an AI reading that listing may well take it at face value, and Amazon itself will be checking your content against reality. Answer those questions properly in your own words, in the listing, and you close a gap your competitors are leaving open.
Say who it's for and who it isn't. A huge part of a good recommendation is fit. A source document that clearly signals "this is the right choice for a frequent business traveller" and, by implication, "not the ultralight option for backpackers" helps the AI recommend you to the right person — which is the person who'll actually be satisfied, buy, keep it, and leave you a good review. Precision about fit isn't turning customers away; it's earning the customers who'll make you look good.
The honest test
Here's the test I'd apply to any listing now: if you handed it, cold, to someone who'd never seen the product, could they use it to confidently answer the ten most common questions a buyer in your category asks? If yes, it's a good source document, and the AI will treat it as one. If they'd be left guessing on half of them, you've got a keyword list wearing the costume of a listing, and the AI will quietly recommend the competitor who wrote proper evidence.
Stop writing to contain words. Start writing to be the reference material — the clearest, most complete, most specific answer to the question your customer is really asking. That's what a source document is, and it's what your listing has become whether you've noticed or not.
FAQ
What's the practical difference between writing for keywords and writing for an AI source document?
Keyword writing aims to contain the right words so an algorithm matches your listing to a search term. Source document writing aims to supply enough specific, checkable facts (dimensions, materials, compatibility, real numbers) that an AI can confidently use your listing as evidence to answer a shopper's actual question.
Where do I find out what questions I should be answering in my listing?
The customer Q&A section Amazon posts on your listing and your competitors' listings, plus review complaints, is a direct feed of what real buyers in your category are unsure about. Just verify the answers yourself before writing them into your copy, since Q&A answers come from other customers and aren't always correct.
Does writing for AI discovery mean my listing should try to appeal to everyone?
No, the opposite. Clearly signalling who a product is and isn't for helps an AI recommend it to the right person, the one who'll actually be satisfied and leave a good review, rather than diluting the listing trying to be a fit for everyone.
About the author
Zamir Cajee is co-founder of This Way Up, a UK business specialising in Amazon marketplace strategy, and co-host of The Upside Podcast, where he and the team break down how Amazon actually works — and how it lies to you. Zamir has built multi-million dollar businesses from scratch and has been selling into the EU since 2016.
Ready to take your Amazon Store to the next level? Start by exploring the tools available to create a standout Storefront or make the process even easier by reaching out to us for tailored guidance.
Subscribe to the This Way Up newsletter for impactful strategies and insights, or tune in to The Upside podcast for expert advice on successfully managing your e-commerce efforts.