A worked example: what to fix first on a 500-product online shop
The shop on this page is made up. We do not publish customer names, customer numbers, or customer quotes.
The scoring is real. Every metric name and every weight below comes straight from the GEO-Score engine that runs your analysis.
So the shop is imaginary. The maths is not. Run the same analysis on your own product page and you get real numbers back.
Most product pages carry a spec table and two lines of marketing copy. An AI assistant has almost nothing to summarise. It picks a competitor page that explains more.
Shops often reuse the manufacturer text on hundreds of pages. AI engines see near-identical pages and cannot tell which one deserves the citation.
People ask AI assistants practical questions. Will it fit a small room? How do I clean it? A page that answers those questions gets recommended. A page that only lists dimensions does not.
Many shop platforms ship a default robots.txt that blocks unknown user agents. If GPTBot or PerplexityBot cannot fetch the page, nothing else on this list matters.
Picture an online shop selling home furniture. It has around 500 product pages, a handful of category pages, and no blog.
Traditional search works fine. The shop ranks for its product names. But when someone asks an AI assistant for a sofa recommendation, the shop is never mentioned.
The owner runs a GEO analysis on one representative product page. The overall score comes back low, and the report shows which of the 22 metrics pulled it down.
That last part is the useful bit. A single overall number tells you little. The per-metric breakdown tells you exactly where the missing points are, and how many of them there are.
| Metric | Share of total score | What to do on a product page |
|---|---|---|
| AI Optimization | 10% | Write a short, direct answer near the top of the page. State what the product is, who it is for, and what makes it different. |
| Comprehensiveness | 8% | Replace the two-line description. Cover materials, sizing, care, delivery, and a realistic use case. |
| Content Structure | 8% | Use one H1 for the product name. Use H2 headings for specifications, care, and delivery. Keep paragraphs short. |
| Answer Completeness | 6% |
Those seven metrics are worth 44 points of the 100 available. That is not a projection. It is how the score is weighted.
The other 56 points sit in metrics that are harder to move on a product page, such as topical authority and knowledge graph presence. Those need site-wide work over months.
So the order below is not a guess. It follows the weights, cheapest fixes first.
Check robots.txt first. If AI bots are blocked, every other improvement is invisible to them. Worth 5% of the score and usually one line of config.
Pick your best-selling product. Rewrite the description, add headings, answer the top questions, fix the alt text. Re-analyse it and see what moved.
Once one page scores well, you know the recipe. Convert it into a content template your team fills in for every product.
Start with the products that make you money. A hundred well-written pages beat five hundred half-written ones.
Never publish the manufacturer description unchanged. If ten shops carry the same text, no AI engine has a reason to cite yours.
Write the answer before the sales pitch. AI assistants quote the sentence that answers the question, not the one that sells.
Out-of-stock pages still get crawled. Keep the content live and mark availability in schema instead of hiding the page.
Category pages are cheap wins. There are far fewer of them, and they carry the buying-guide content that builds authority.
Re-analyse after every batch. A metric that will not move usually points at a template problem, not a content problem.
How the engine decides whether a page covers its topic properly. Worth 8% of the total score.
Headings, paragraph length, and lists. Worth 8%, and the easiest big metric to fix on a product template.
Which crawlers can reach your pages, and how to check. Worth 5% and usually a config fix.
Writing alt text that describes the product instead of the file. Worth 4%.
Analyse one real product page and see your own per-metric breakdown. No invented numbers, just your page.
Analyse a page| Add the five questions your support team hears most, with real answers. Not a generic FAQ block. |
| AI Bot Access | 5% | Check robots.txt allows GPTBot, ClaudeBot, PerplexityBot, and Google-Extended. This is usually a five-minute fix. |
| Image Alt Text | 4% | Describe what the photo shows. "Grey three-seat sofa in a living room" beats "sofa-1.jpg". |
| Schema Validator | 3% | Add valid Product schema with price, availability, and reviews. Then test it, because broken schema scores the same as no schema. |
A guide covering a whole category builds topical authority. This is the slow metric, so start it early and let it mature.