When a buyer asks AI what to buy, your brand isn’t in the answer.
Answer engines recommend what they can parse. We test whether yours can be read at all — crawl permission, product data, and live presence in the answers — then fix the specific defects. 14 of 22 mid-market brands we scanned last week publish no price a machine can read.
A score and a defect list in 24 hours. No call required.
AI-referred traffic to retail sites, year over year. Adobe, Sept 2026.
US online spend this holiday season, +6.7%. Adobe, Sept 2026.
Of 22 mid-market brands scanned: no product schema on a real product page. Our scan, Oct 2026.
Brands whose robots.txt still forbids every major AI crawler outright.
Four things decide whether you can be quoted.
Not a vibe check on “AI readiness.” Four measurable gates, each pass or fail, each with a URL attached.
Crawl permission
Which AI agents your robots.txt actually admits. GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, CCBot. A blocked bot cannot cite you, whatever your traffic report says.
Product data
JSON-LD Product / Offer / ProductGroup on real product pages — name, price, availability, currency. This is the layer answer engines quote from.
Answer presence
25 real buying prompts per category, run live against the major assistants, screenshotted. You either appear or you don’t — and we show you what the buyer was told instead.
Competitor gap
Who gets named when you don’t, and the specific markup difference that explains it. That comparison is the business case, not an opinion.
The gap is usually in the source code.
A mid-market home brand with healthy organic traffic. Its product pages carry zero product markup. Three competitors publish price and stock on every page. When a shopper asks an assistant for a recommendation in that category, the brand is never in the answer — and nothing in its analytics says why.
Scanned free. Priced by how much of the catalog is broken.
Same probe at every size. The work scales with SKU count and how many prompts you want tracked.
- Full visibility score, 0–100
- Crawl, schema, llms.txt defects with URLs
- Product + Offer markup shipped on the catalog
- Re-scan and before/after proof
- Everything above, whole catalog
- 50-prompt panel across your categories
- Competitor share-of-answer table
- Feed parity check against your channels
- 30-day re-test included
- 200-prompt panel, per category
- Share-of-answer versus named competitors
- Monthly re-test and exec readout
- Engineering-ready defect tickets
If the scan finds nothing material, you don’t pay for the scan. If the fix doesn’t move your measured share of answer in 60 days, you don’t pay the second half. You see the defects and the evidence before any work starts.
The four that always come up.
Isn’t this just SEO?
No. SEO optimizes for a ranked list of links a human clicks. This optimizes for being the named answer inside a paragraph a machine writes. Different crawlers, different parsing, different success metric — share of answer, not position.
What exactly do you change?
Product and Offer structured data on every product page, crawl permissions, an llms.txt summary, feed consistency, and the factual descriptions the models quote. All of it is on your own site, none of it touches your design.
How do you prove it worked?
The same 25–200 prompts, re-run, screenshotted, side by side. Same prompts, same models, same day of week. You can re-run them yourself at any time — the list is yours whether you hire us or not.
Can this be live before Black Friday?
Yes for a single store: 48 hours. Catalog work is 10 days. Anything after Nov 20 is post-peak — the holiday traffic will already have been routed by then.
Find out what the machines can’t read.
Your domain in, a score and a defect list back within 24 hours. If there’s no gap, we’ll say so and you’ll have a clean bill of health in writing.