GEO (Generative Engine Optimization) is the practice of getting your content chosen as a source when an AI assistant builds an answer. For a consumer brand that means something very specific: when someone asks what to buy in your category, AI names you, describes you correctly and sends people to buy somewhere that works for you. And the audience is no longer niche: 34% of US adults say they have used ChatGPT, roughly double the share in 2023 (Pew Research Center, June 2025).
Most GEO guides are written for software or services companies. In consumer goods and retail the game is different for one reason: much of what AI reads about your product isn’t on your site. It’s on your Mercado Libre, Amazon, Walmart and Liverpool listings, in your shoppers’ reviews and in media you never hired.
These are the 9 actions that move the needle most, ordered from quickest to most structural.
1. Measure before you touch anything
Without a baseline there’s no way to know whether anything worked. Build a set of 20 to 30 real shopping questions, run them on ChatGPT, Gemini, Perplexity and Google AI Mode with clean accounts, and log who shows up, in what order and which sources are cited. We explain it step by step in AI Shelf: what it is and how to measure whether ChatGPT recommends your brand.
2. List the sources AI cites in your category
This is the most actionable finding of any measurement. If Perplexity and ChatGPT keep citing the same comparison site, the same nutrition blog or the same Amazon listing, that’s where you need to be well represented. Sometimes the best GEO action isn’t publishing more on your site, but fixing a listing at a retailer.
3. Answer the exact question, in the first line
Assistants extract snippets. A page that opens with “At [brand] we believe in quality nutrition” helps no one. One that opens with “Recommended for children ages 1 to 10 with low appetite; 2 servings a day” can actually be cited.
For each purchase intent where you want to show up, create a section or page that:
- opens with the direct answer in one or two sentences;
- includes the fact that makes it verifiable (age, dose, content, certification);
- compares honestly against the alternatives.
4. Fill in the attributes on every retailer listing
AI compares products based on attributes: size, flavor, nutritional content, sugar-free, suitable for, pack size. A listing with half its attributes empty loses to a complete one, even if the product is better. It’s the same work the Digital Shelf demands, and now it’s worth twice as much.
5. Unify data across your site, retailers and marketplaces
When price, pack size or a claim changes depending on the source, AI mixes up data or drops the product. A study by Facttory in Spain found that more than 15% of products were described inaccurately by the models, and attributed much of it to contradictory information between brand websites and their retail partners (Ecosistema Startup). Take an inventory: name, pack size, suggested price and main claims, identical everywhere.
6. Work on reviews and questions on marketplaces
Reviews are AI’s favorite source for “users highlight…” arguments. Three concrete actions:
- answer every open question on your main listings, with data;
- turn on verified review programs where the marketplace allows it;
- read the negative reviews: if a complaint repeats, AI will repeat it.
7. Implement structured data and an llms.txt
On your site, Product schema (with Offer, price, availability and brand), Organization and FAQPage help systems understand what each thing is. An llms.txt file at the root, with a short description of the brand and links to your key pages, is cheap and organizes how you’re read. It doesn’t work miracles, but it removes friction.
8. Get third parties to talk about you with data
AI trusts what others say more than what you say. The pieces that work best for a consumer brand:
- studies with proprietary data that media can cite;
- presence in comparisons and buying guides from specialized media;
- expert answers (nutritionists, dermatologists, technicians) with name and credentials.
9. Measure every month and act on the biggest gap
GEO isn’t a project that ends. Every model version changes the answers, and so does every move your competitors make. Each month, pick the biggest gap (an intent where you don’t appear, a source that describes you wrong, a path to purchase that goes to a third party) and work on that one only.
What doesn’t work
It needs to be said, because a lot of it is being sold:
- Stuffing pages with “best X in Mexico” keywords. AI doesn’t count repetitions; it looks for verifiable answers.
- Publishing hundreds of generic AI-generated articles. They dilute your domain’s authority and are almost never cited.
- Promising to “show up first in ChatGPT” in 30 days. Nobody controls a model’s answer. What you can control are the sources and content the model reads.
Where to start this week
If you can only do one thing, do action 2: find out which sources AI cites when it talks about your category. That list tells you where to invest for the next three months.
At Balloon we include it in the free AI Shelf and Digital Shelf assessment: we compare your brand with three competitors, list the sources AI uses and give you three actions with estimated impact. Request it here.
FAQ
Does GEO replace SEO?
No. AI assistants use search engines to find sources, so a solid SEO foundation is still a requirement. GEO adds a layer: making your content easy to cite and getting your brand into the sources AI checks.
How long does it take to see a change?
Assistants with live web search can reflect changes within weeks. Anything that depends on model training can take months. That's why it pays to measure monthly and not expect overnight results.
Do I need an llms.txt file?
It isn't a mandatory standard, but it's cheap to create and helps AI systems understand what your brand is and which pages matter most. We recommend it as part of technical hygiene.