AI Personalization Evaluation Expert: Multilingual

What you'll do

You test how personal an AI assistant can get. Each session starts with you writing a prompt, then holding a natural back-and-forth with two versions of an assistant. Some of your questions can touch information or activity from the digital services you have connected; others can be anything on your mind.

After each conversation you rate both experiences. The listing names the dimensions: whether the assistant used the available information well, felt like it understood you and your context, connected relevant information across the conversation, gave trustworthy and safe answers, handled your feedback and corrections, and felt helpful and natural overall. It is a side-by-side comparison, so the useful skill is pointing to concrete differences between the two.

The eligibility check

Entry to this project depends on your real usage. It needs a certain threshold of recent activity across a specific combination of commonly used digital services, and a short verification checks that before you can take part. The listing does not name the services or the threshold, and says different combinations of qualifying activity may be considered, so answer from what you actually use now. Not qualifying does not rule you out of other Invisible evaluation work, according to the listing.

Because the assistant works with information from services connected to you, the listing does not say what access is granted, what data the assistant sees or how it is retained. Check the Statement of Work before connecting anything.

Languages

The listing runs as ten tracks, one per language or locale:

• Spanish (Latin America)
• Arabic (Global)
• Portuguese (Brazil)
• Japanese (Japan)
• Turkish (Turkey)
• Indonesian (Indonesia)
• Hindi (India)
• Korean (Korea)
• French (France)
• German (Germany)

You need fluency in one of them, plus the matching expert requirement set, which the listing does not spell out. The locale matters because personalization is local: an assistant that handles Brazilian slang, Japanese politeness levels or Indian date formats well in one conversation and poorly in the next is exactly what the ratings are meant to catch. English is not listed as a working language.

What it pays

$17 an hour, the same for every language track. That is the standard rate for Invisible's general evaluation work, and the same as the Image Data Contributor and Evaluator listing, which tests a similar idea with your photos instead of your accounts.

The contract terms say payment is weekly via Stripe or Wise, and that Invisible cannot support H-1B or STEM OPT contractors.

Worth knowing

Good:

• Conversational work that needs no specialist background
• Covers languages that often have fewer evaluation openings, such as Turkish, Indonesian and Hindi
• Remote, on your own schedule

Less good:

• $17 an hour, low for the market
• Eligibility depends on your personal service usage, which you cannot easily change
• The listing does not name the services or say how connected data is handled
• A verification step on top of the resume upload (about 5 minutes) and unpaid qualification step (about 30 minutes), with no guarantee of placement
• No minimum amount of work; independent contractor terms with no benefits

About this listing

Read on the signed-in Meridial marketplace on 26 September 2026, posted about a month before we read it. The tasks, assessment dimensions, eligibility rules, language tracks, rate and contract terms are the listing's own; the notes on data handling and locale are ours. For a photo-based counterpart see Image Data Contributor and Evaluator. See Invisible Technologies.