Prediction ledger / Study 1
Which Christmas ads will Britain like best? AI answers first.
On the day each UK Christmas ad comes out, three AI models predict how viewers will rate it. We publish only a fingerprint of the prediction that day. After System1’s final Christmas ranking we open every prediction and score it by rules fixed in advance. Anyone can check that nothing changed in between.
On this page
The ledger
| Sealed | What | Fingerprint (SHA-256) |
|---|---|---|
| 2026-10-29 10:52 | Protocol locked | b0f193f614141e6e57be9580de738d99dc78cded0fef5ec6b824eaffa107a873 |
| 2026-10-29 10:56 | Dr. Oetker, “Share a little Joy this Christmas” | 2f8b08e9098b88fbdb9d721edbe41a95ce2043e6a3ba84c340976e19394234df |
| 2026-10-29 10:56 | Debenhams, “Christmas. For Everyone” | 955dc9ccb23714d4cd79e0125f6fb513af26d87d7b98649030c8bbb8cefa4b11 |
| 2026-10-29 10:57 | The Perfume Shop, “Let Love Last” | e9152f5a485b5c371a195daec14bcb6e69d5ac8742633bb8b1d85be2bcb44696 |
| 2026-10-30 14:54 | Pandora, “More than a gift” | ca0245b818f3ff5d6b14e05863897f962dd5594ba4906a0127778c04ac811fc9 |
| 2026-11-01 10:09 | Tesco, “Because this Christmas, every little helps” | 3c8bf18b3e24caeff9f4659b9b5fdf84a872098261c25448b0e3cf4daaf9981f |
| 2026-11-01 10:11 | Sainsbury's, “Sainsbury’s big delivery.” | 596c5b4e9d9b41c92a9981508928dfb70f3f8d5e6ce1aaaf47b210c369de0a7b |
| 2026-11-01 10:12 | Argos, “Argos Christmas Advert 2026” | 1ad92291d6ce0683e5b67efdfc8e958f58e1ed10fc235504314e7fcc232cbfb7 |
How it works
- Gemini watches the ad through its official YouTube link and writes a scene-by-scene description. A second Google model, which does not predict, removes any judging words. We do not download the video.
- Models from OpenAI, Google and Anthropic read the same description and predict the ad’s System1 Star Rating, ten times each, with web search off. Gemini also predicts from the video itself.
- The same day we publish the SHA-256 fingerprint of the full record here and on X, and stamp it with OpenTimestamps, a free public timestamping service. The record stays sealed.
- Within seven days of System1’s final Christmas ranking, expected in late November, we open every record and score it.
The rules
The protocol fixes the ads, the exact prompts, the metrics and what counts as a hit before the first seal. The study script reads its prompts from the protocol and refuses to run if the file has changed. A sealed prediction is never re-run, and any change to the rules gets a new fingerprint and a line in the ledger.
Locked on 2026-10-29 10:52 UK time. Fingerprint: b0f193f614141e6e57be9580de738d99dc78cded0fef5ec6b824eaffa107a873. Read the protocol, the exact file we fingerprinted.
Brands we follow
We seal every UK Christmas ad these brands release between 15 October and 5 December 2026. The list is never shortened.
- Airbnb
- Aldi
- Amazon
- Apple
- Argos
- Arsenal
- Asda
- BBC
- Baileys
- Barbour
- Boots
- Burberry
- Cadbury
- Cartier
- Chelsea FC
- Coca-Cola
- Debenhams
- Disney
- Disney+
- Dogs Trust
- Dr. Oetker
- EE
- Etsy
- Fat Face
- Fortnum & Mason
- Freemans
- GAP
- Good Housekeeping
- Google Pixel
- Greggs
- Harvey Nichols
- Heathrow
- Hotel Chocolat
- JD Sports
- John Lewis
- KFC
- LEGO
- Lidl
- Lindt Lindor
- M&S Clothing & Home
- M&S Food
- McDonald's
- Moonpig
- Morrisons
- National Trust
- Nespresso
- Ninja
- O2
- Pandora
- People's Postcode Lottery
- Sainsbury's
- Sanpellegrino
- Save the Children
- Sky
- Smyths
- Sports Direct
- Starbucks
- TK Maxx
- TUI
- Tesco
- The Perfume Shop
- Trafford Centre
- Treasure Tress
- Very
- Vodafone
- Waitrose
Seal your own prediction
Anyone with a way to predict how ads will be rated, including testing companies, researchers and AI developers, can seal predictions for the same ads under the same rules. Send the SHA-256 fingerprint of your prediction record to hello@kesterley.com before midnight UK time on the ad’s release day, with a random string of at least 16 characters inside the record so the fingerprint cannot be guessed. We list every fingerprint with the time it arrived and, after System1’s final ranking, score every opened record by the same rules as ours. Late fingerprints are listed but not scored.
Check a fingerprint
After the opening, take any published record, remove its sha256 field, write the rest with sorted keys and no spaces (RFC 8785) and hash it. In Python:
import hashlib, json
record = json.load(open("prediction.json", encoding="utf-8"))
sealed = record.pop("sha256")
canonical = json.dumps(record, sort_keys=True, separators=(",", ":"), ensure_ascii=False)
print(hashlib.sha256(canonical.encode("utf-8")).hexdigest() == sealed)Who pays for this
Nobody paid for this study. No brand, testing company or AI company saw it in advance. From System1’s ranking we cite only what System1 publishes, with the source.