Core dna (@coredna) header
Core dna (@coredna) profile picture

Core dna

@coredna

Modern DXP for the Enterprise

 Global
coredna.com/
Joined October 2009
  • Tweets10.4K
  • Following3.3K
  • Followers3K
5.1KMedia
Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)Photo by Core dna (@coredna)
Black Friday check for retailers: in our ChatGPT test, Kogan First and EB Games PLUS members’ prices were quoted as the price everyone pays. Put the condition next to the price on the page, in your feeds and in every deal you brief out. Give every offer an end date too.
Warm beige checklist card titled Before Black Friday: conditions next to the price, end dates on every deal, feed updates in the promotion calendar, brief your service team.

ALT Warm beige checklist card titled Before Black Friday: conditions next to the price, end dates on every deal, feed updates in the promotion calendar, brief your service team.

1
17
The full timeline for the weeks before Black Friday: coredna.com/blogs/chatgpt-bl…
ChatGPT Black Friday Test | Analyzing AI's Pricing Accuracy | Core dna

ChatGPT Black Friday Test | Analyzing AI's Pricing Accuracy | Core dna

ChatGPT's pricing accuracy for Black Friday products was tested. 21 out of 40 prices were found incorrect, with all errors showing prices too low.

coredna.com
4
Target now feeds what shoppers tell its AI assistant back into its website and app. An Italian food preference from one chat was stored at 92% confidence, time-bound, never as a permanent fact. If chat data vanishes when the session ends, your site never learns from it.
Lime card. Large text: 92% confidence. An Italian food preference from one assistant conversation, time-bound, sourced, never permanent. Target, RecSys 2026.

ALT Lime card. Large text: 92% confidence. An Italian food preference from one assistant conversation, time-bound, sourced, never permanent. Target, RecSys 2026.

1
14
How Target, Google and Spotify are changing recommendations: coredna.com/blogs/recommenda…
Recommendation Engines | How They Work, Types and Examples | Core dna

Recommendation Engines | How They Work, Types and Examples | Core dna

Discover how 80% of Netflix viewing comes from recommendations. Learn how recommendation engines function, their main types, and their impact on eCommerce.

coredna.com
3
With Core dna MCP 2.0 you can type this into Claude or ChatGPT: “Create a landing page featuring our diamond rings at 25% off. Launch it on Black Friday and restore regular prices after Cyber Monday.” The agent builds it. Deletes still need your OK.
Illustration of a chat message on a pale blue card: Create a landing page featuring our diamond rings at 25% off. Launch it on Black Friday and restore regular prices after Cyber Monday.

ALT Illustration of a chat message on a pale blue card: Create a landing page featuring our diamond rings at 25% off. Launch it on Black Friday and restore regular prices after Cyber Monday.

1
20
5 more prompts to try: coredna.com/blogs/whats-new-…
What's New in Core dna MCP 2.0: Automations, New Content Types and More | Core dna

What's New in Core dna MCP 2.0: Automations, New Content Types and More | Core dna

Learn about Core dna MCP 2.0's capabilities, including AI-driven automations, new content types, and page editing enhancements.

coredna.com
6
JB Hi-Fi appeared in 21 of 40 ChatGPT answers in our Black Friday test. It was the recommended store in 2. Officeworks appeared in 13 and was recommended in 7. Only the recommendation sends the shopper to your site. Track named rate and recommended rate as two numbers.
Bar chart on a light card, named versus recommended in 40 ChatGPT answers. JB Hi-Fi named 21, recommended 2. Officeworks named 13, recommended 7.

ALT Bar chart on a light card, named versus recommended in 40 ChatGPT answers. JB Hi-Fi named 21, recommended 2. Officeworks named 13, recommended 7.

1
33
The four measures we’d track before Black Friday: coredna.com/blogs/chatgpt-bl…
ChatGPT Black Friday Test | Analyzing AI's Pricing Accuracy | Core dna

ChatGPT Black Friday Test | Analyzing AI's Pricing Accuracy | Core dna

ChatGPT's pricing accuracy for Black Friday products was tested. 21 out of 40 prices were found incorrect, with all errors showing prices too low.

coredna.com
6
Margaret Mitchell of Hugging Face, opening RecSys 2026: “an approving click doesn’t necessarily indicate the best action.” Tired reviewers approve fluent plans, so agents learn to produce what is easy to approve. Her test: plant a known error and see if it gets caught.
Brown card with a quote: Design what agents show their users by value, not by approval rate. Margaret Mitchell, Hugging Face, RecSys 2026 keynote.

ALT Brown card with a quote: Design what agents show their users by value, not by approval rate. Margaret Mitchell, Hugging Face, RecSys 2026 keynote.

1
12
Her full argument, and how to measure AI properly: coredna.com/blogs/recommenda…
Recommendation Engines | How They Work, Types and Examples | Core dna

Recommendation Engines | How They Work, Types and Examples | Core dna

Discover how 80% of Netflix viewing comes from recommendations. Learn how recommendation engines function, their main types, and their impact on eCommerce.

coredna.com
2
Our CTO gave Claude one brief: live reporting for a client’s sign-up forms, some with more than 300,000 submissions. Claude wrote a plan. Dmitry reviewed it. Then Claude built a new content type, a 13-step automation and a 26-step automation on a dev copy of the site.
Illustration on a beige card: Brief, then Plan reviewed by the CTO, then Built: a new content type, a 13-step automation and a 26-step automation.

ALT Illustration on a beige card: Brief, then Plan reviewed by the CTO, then Built: a new content type, a 13-step automation and a 26-step automation.

1
66
How the build worked, step by step: coredna.com/blogs/whats-new-…
What's New in Core dna MCP 2.0: Automations, New Content Types and More | Core dna

What's New in Core dna MCP 2.0: Automations, New Content Types and More | Core dna

Learn about Core dna MCP 2.0's capabilities, including AI-driven automations, new content types, and page editing enhancements.

coredna.com
1
Before 1 November, ask ChatGPT about your top Black Friday products the way a shopper would, in a fresh chat. Check the price, then check which site the answer cites. A right price from a comparison site is the one that won’t change when your sale starts.
Image from Core dna (@coredna)
1
20
15 copy-and-paste prompts to run the same test on your own products: coredna.com/guides/prompts-a…
15 copy-and-paste prompts to find if ChatGPT can find your products | Core dna

15 copy-and-paste prompts to find if ChatGPT can find your products | Core dna

15 copy-and-paste prompts to find out whether AI can actually find, understand and recommend what you sell, with a scorecard and a map from each failure to the layer responsible.

coredna.com
9
When ChatGPT recommended one of Australia’s most-loved retailers in our Black Friday test, the price was wrong in 15 of 24 cases. Every wrong price was lower than the store was charging. The retailers’ own pages were right. The answers took their numbers from elsewhere.
Deep red card. Large text: 15 of 24 Black Friday recommendations had the wrong price. Every wrong price was too low. Example: Garmin Forerunner 265, ChatGPT said $479, store charged $797.

ALT Deep red card. Large text: 15 of 24 Black Friday recommendations had the wrong price. Every wrong price was too low. Example: Garmin Forerunner 265, ChatGPT said $479, store charged $797.

1
14
Where the wrong numbers came from, source by source: coredna.com/blogs/chatgpt-bl…
ChatGPT Black Friday Test | Analyzing AI's Pricing Accuracy | Core dna

ChatGPT Black Friday Test | Analyzing AI's Pricing Accuracy | Core dna

ChatGPT's pricing accuracy for Black Friday products was tested. 21 out of 40 prices were found incorrect, with all errors showing prices too low.

coredna.com
3
About 70% of experiments on YouTube’s watch page ranking model are now run by an AI system, YouTube said at RecSys 2026. It found a new reward factor no engineer had considered, and it shipped. The catch: left alone, the agents played it safe. They had to be told to explore.
Dark card. Large text: about 70% of experiments on YouTube’s watch page ranking model are run by an AI system. Source: YouTube at RecSys 2026.

ALT Dark card. Large text: about 70% of experiments on YouTube’s watch page ranking model are run by an AI system. Source: YouTube at RecSys 2026.

1
15
What else we learned at RecSys 2026: coredna.com/blogs/recommenda…
Recommendation Engines | How They Work, Types and Examples | Core dna

Recommendation Engines | How They Work, Types and Examples | Core dna

Discover how 80% of Netflix viewing comes from recommendations. Learn how recommendation engines function, their main types, and their impact on eCommerce.

coredna.com
6
We also asked whether each product was in stock near Richmond, Victoria. None of the 48 stock answers we could score confirmed a product in a specific store. Where the retailer is silent, someone else answers: one pet retailer’s stock came from its Uber Eats listing.
Image from Core dna (@coredna)
1
20
The stock results in full, including three in-stock products reported as unavailable: coredna.com/blogs/black-frid…
Why ChatGPT Sometimes Shows Wrong Prices: 51-Retailer Test | Core dna | Core dna

Why ChatGPT Sometimes Shows Wrong Prices: 51-Retailer Test | Core dna | Core dna

We checked ChatGPT's price and stock answers for 51 Australian retailers against their live product pages. Every wrong price came from another page.

coredna.com
8
Load more