12. Error examples
In Spanish translations, ChatGPT was able to produce readable,
grammatically correct text. However, some of the nuances of local
phrases were lost and the output felt ‘plain’ and lacking emotion.
In French, the AI translations were readable but the
arrangement of the words stuck too closely to the original
English source, leading to rigid text that didn’t flow well.
ChatGPT had the most problems with Chinese translation,
including mistranslations and broken phrases, along with
a lack of local audience considerations.
ES
FR
ZH
13. Editorial thoughts
For EU markets it produced readable
and grammatically correct text
Editors felt the AI copy was fairly
obviously machine translated
Mistakes - major and critical
The AI translation felt flat vs human -
lacking an emotive connection
14. Survey of AI trust
Surveyed 2,500 people with a
familiarity with AI
500 per market in the UK, US,
France, Spain, and China
15. How do you feel about companies using AI to write communications
19. This is positive, right? Well, it depends how you think…
40%
Either on the fence, negative or very negative
feeling towards AI being involved in YMYL
content
20. And it depends on the market…
48%
When you take China out of the stats
24. Survey thoughts
Around 60% positivity for most
content types
Big variations between China
and Europe/US
Product descriptions around less
important topics could be OK
26. Low quality on-site copy
Meta descriptions
Product descriptions
Creative blogs
ChatGPT will raise standards, but also lower them
Depending on tone of
voice and how niche the
product is
Thought
leading
32. Closing thoughts
EXPERIENCE and EXPERTISE: There will
be an overreliance on AI soon
AUTHORITY: For those top positions on
Google, it could hinder rather than help
TRUST: Consumer perception is fairly
positive, but also some skepticism
● Don’t assume time saving
● Plenty of amazing applications for AI
around creative copy
Detail what localisation is and how it differs from straight translation - why might you use it in this context? Data first approach
Just an intro slide - bit of a joke about its awful name and assumption most people know. Touch on its more advanced language capabilities vs previous version
Bringing them together to test effectiveness for longform editorial
Thinking always about Google’s measurement framework
We ran two tests
Starting with editorial
Our scoring criteria
Our scoring criteria
AI didn’t do as well
More errors - detail specifics
Wrap up notes
Onto the survey
Mostly fairly positive outlook
Less confident with longform editorial
Less confident with longform editorial
Less confident with longform editorial
Onto the survey
Detail what localisation is and how it differs from straight translation - why might you use it in this context? Data first approach