christmasornot

In late March, the founder of Christmasornot ran our free scan on a whim. The result was a 24 — bottom-decile, labeled "Critical." ChatGPT could not state the company's pricing. Gemini placed the founding year four years too early. Perplexity confused the product with an unrelated holiday e-commerce site. Three months earlier they'd raised a seed round. The pitch deck said "we are the only AI holiday classifier on the market." The models had no idea.

What followed is, by AEO standards, a textbook recovery — measured, methodical, executed in-house in under thirty days. We're publishing it with permission and with the receipts. The before-and-after reports are linked at the bottom.

Chapter 01The problem

Christmasornot is a four-person team in Brooklyn that ships a single product: an API that returns whether a given image, song, or paragraph "feels Christmassy." It's used by ad agencies for seasonal targeting and by a handful of e-commerce platforms for category routing. The product is good. The brand surface was not.

Their initial scan surfaced six failure modes, in roughly descending order of severity:

• Pricing was unanswerable. Their pricing page used "Contact us" boxes for all three tiers. No schema. No machine-readable price.
• Founding date was wrong. An old TechCrunch piece misstated the year. Models had latched onto it.
• Customer list was hallucinated. Their actual customers were named in PDF case studies. Models couldn't read them.
• Product category was ambiguous. "AI classifier" + "Christmas" parsed as either "AI ornament shop" or "Christmas movie recommender" — depending on the model.
• Founder bio was buried. The about page had a hero video; no plain-text bio. Models couldn't quote it.
• No llms.txt, no JSON-LD, no Organization markup. The basics weren't shipped.

Chapter 02What we found

The Free scan caught the surface failures. The $49 Full Consensus Report exposed the structural ones. Each of the six findings had a fix; the team prioritized by leverage. We helped them sort.

The leverage analysis was unsentimental: pricing-schema and llms.txt would move four of the six failures by themselves. The customer list and founder bio could wait until those landed. The TechCrunch correction would happen through a press follow-up — a slower lever, but worth pulling.

Chapter 03What they changed

Week 1 — Schema overhaul

They shipped Organization, Product, and FAQPage JSON-LD across the site. The Product schema included the actual prices — $19, $49, $199 — with proper priceCurrency and priceValidUntil. They removed the "Contact us" boxes from the pricing page and replaced them with a plain table.

Week 2 — llms.txt + content rewrite

They published a 40-line llms.txt with a tight summary block, clear section headings, and links to their docs, pricing, customers, and about pages. The about page got rewritten as plain text under the hero video. The customer logos got plain-text alt attributes and machine-readable JSON.

Week 3 — Press + sources cleanup

The founder reached out to the TechCrunch reporter and got a correction note appended. They updated their Crunchbase, LinkedIn, and Wikidata entries with consistent founding-date data. This is the slowest-moving lever — and it pays the longest dividend.

Week 4 — Re-scan + verify

They ran a fresh scan on day 28. The score had moved from 24 to 68. They shipped two more rounds of schema tweaks. By day 32 the score read 78.

Chapter 04The results

By the end of the month, all three models could correctly state Christmasornot's pricing, founding year, customer list, and product category. The consensus matrix moved from "all three disagree" to "all three agree, with citations" on five of the six original questions. The sixth — a question about Series A funding that hadn't happened yet — correctly returns "no Series A on record" from all three models.

Chapter 05What stuck

Christmasornot upgraded to Monitor Pro three weeks after the re-scan. The reason was simple: models drift. GPT-4.5 shipped in mid-April and briefly returned them to a lower agreement score until the index caught up. They saw it within 48 hours of the model release because their weekly scan caught it. The team made one schema tweak and the score returned. Without monitoring, they wouldn't have known.

That's the operational reality of AEO at the brand level: it's not a one-time fix. The models keep moving. Your facts stay still. The gap between "true" and "what AI says" opens back up unless something is watching.