How to get cited by AI
When a shopper asks AI what to buy, it cites a few sources. These six steps make your product page one the AI surfaces in its answer.
Open the product page with a clear definitional or claim sentence (who needs this, what it is, why) — and put that same sentence into the structured data as the Product description. AI reads the structured data first, so the sentence belongs in both places, not only in the page text.
Specs, origin, dimensions, use, compatibility. AI cites specific and verifiable detail, not marketing words.
Write the questions a customer actually asks, and answer briefly. These question-answer pairs are easy for AI to cite.
Product + Offer + FAQPage JSON-LD in the page code. It tells the machine what the page contains and makes it easier to cite.
AI answers in the asker's language and cites that language's pages. Content and schema must be in each language page's own language.
Run the real buyer queries in ChatGPT, Perplexity and Google AI and see who they cite. Fix the page accordingly and measure again.
How many fields come back empty?
AI doesn't read your product page the way a human does. It reaches for the structured data and fills its answer from that. This prompt walks your page's text item by item and marks each one present, partial or missing — and surfaces the contradictions. It writes no fix; it tells you what a buyer never gets to know about you.
Do not browse the web. Do not invent anything. Do not fill in missing information. Here is the visible text of one of my product pages: --- --- 1. Go through these items and mark each one with exactly one value: PRESENT / PARTIAL / MISSING. Do not write out the value itself, only the mark. product name · description · product code · GTIN/EAN · manufacturer · material · colour · size · weight · price · currency · availability · return policy · delivery time · reviews · temperature or use rating 2. For every PARTIAL: one sentence on what is missing from it. Do not complete it. 3. Find the contradictions and ambiguities in the text: places where two facts conflict, or where it isn't clear which variant a number refers to. 4. Count: how many PRESENT, how many PARTIAL, how many MISSING. 5. Name three questions a buyer would ask about this product that this data does not answer.
Never paste chat-generated JSON-LD into your store.
Ask any AI to write the schema code for your product page and it reads the values as they are at that second and hardcodes them. Price, availability and stock freeze at that moment. The code is correct for exactly as long as nothing changes.
The first price change, and your structured data tells machines a different price than the page tells a human. Google's guidelines require structured data to match the visible page — a mismatch costs you rich results rather than winning them.
availability stays InStock. A buying agent recommends and tries to purchase a product that doesn't exist. In agentic commerce that isn't a bad search result, it's a failed transaction.
Every price change, campaign and stock movement means generating the code again — for every product, in every language. At four hundred products that isn't laborious, it's impossible.
Structured data belongs in your platform's template, where price, availability and stock are read from the same source as the visible page. Then it stays correct without anyone touching it. That is the work we do — and the reason a pasted snippet doesn't replace it.
- What you saw
- One product's gaps and contradictions. The model couldn't fill them because the information wasn't on the page. The same repeats on every product, in every language — and shifts every time the catalogue does.
- What we do
- We fill those fields across the whole catalogue and put the data where AI reads it. That is work no model can do for you: it doesn't know your GTIN, your material or your return policy.
- Where you start
- The free audit shows what your whole store looks like to a machine. Outcite fixes one product page at a time, with no subscription.
Schema code written by an AI hardcodes whatever values happened to be on the page at the moment it ran. Price, availability and stock freeze at that second, and after the first discount your structured data tells machines a different price than the page tells a human. We put the data into the platform's template, where the values are read from the same source as the visible page — so it stays correct without anyone touching it.
The prompt tells you which fields are missing. It doesn't know their values. The model doesn't know your GTIN, your fill power, your certification or your store-level stock. Ask it to fix the page and it will invent them — and invented product data, once published, is worse than missing data: it's a consumer-law problem and it comes back as returns.
In a chat window you fix one product once. The problem is hundreds or tens of thousands of products, in every language, and again every time the catalogue, the price or the stock changes. That's a pipeline, not a prompt.
The model prints JSON-LD as text in a chat. It still has to live in your platform's template, per language, valid, and survive theme updates. That is wiring work into your platform, not copy-paste.
The model assesses itself from memory, on one AI surface, at one moment — it usually says so itself. We run ChatGPT, Perplexity and Google AI, on the same queries, repeated over time.
The terms used in this field
- GEO
- Generative Engine Optimization — shaping content and data so AI answers cite you.
- AEO
- Answer Engine Optimization — nearly the same as GEO, framed around answer engines.
- AI Overviews
- Google's AI answer above the results; already answers over 2 billion users a month (Google, 2025).
- Sitaatti / citation
- When AI names or links a source in its answer. The goal: to be among the cited.
- schema.org / JSON-LD
- Machine-readable structured data in the page code (e.g. Product, FAQPage) that tells machines what the page contains.
- hreflang
- A tag that declares a page's language versions. AI cites the version in the asker's language.
GEO (Generative Engine Optimization) means shaping content and data so AI answers (ChatGPT, Perplexity, Google AI) surface and cite you. AEO (Answer Engine Optimization) is nearly the same, framed around answer engines. Both aim at the same thing: getting named in the AI's answer, not just the results list.
Not on its own. Ranking in results doesn't mean you get cited in the AI's answer — they are different machinery. AI surfaces sources by how clearly and structurally a page answers the question, not by ranking alone.
A monitoring tool shows a dashboard: where you rank and who gets mentioned. It changes nothing on your page. A fix gives you a concrete deliverable — better page copy and schema code — that you add to your page so AI starts citing you.
Yes. AI answers in the language it's asked and cites that language's pages. The JSON-LD and content must be in each language page's own language to be cited in each market.
Some tools are platform-agnostic, some are locked to a stack. Structured data (schema.org) itself works on every platform; the difference is whether the tool gives you the finished code and steps for your platform, or just a metric.
What does your store look like to AI?
A free audit. No signup, works on every platform.