AI Enrichment

Enrich is the second stage of the pipeline (Collect → Enrich → Deliver). After Collect brings in the products from the supplier, Enrich uses AI to rewrite the description of each one to match your brand's standard, before publishing.

What Enrich does

For each product, the AI receives the name, the cost price, the original description, the spec sheet, and the EAN, and returns a rewritten description — with SEO, storytelling, and technical structure. Products are processed in batches of 3, in parallel.

Activate the module

Enrich runs when the AI Enricher module is active:

• Included in the Flex, Advantage, and Enterprise plans.
• On the Trial plan, it can be purchased separately in Billing → Modules.

Without the module, products go straight to publishing with the supplier's original description — nothing is sent to the AI.

Brand tone of voice

In Settings → Enrichment you define how the AI writes. There are two fields, saved as the account default (only the account owner can edit them):

Brand description prompt

This is the system instruction that guides each rewrite. Click one of the templates to use as a starting point and adjust the text:

Use {brand_name} in the text — it's replaced with the Brand name of each supplier at enrichment time. This way the same prompt works for multiple brands.

Image background style

The prompt for the background swap done by Image AI (stone background, white marble, linen and wood, white studio, or your own). This is also an account default.

If you don't configure anything, Trinis uses a built-in default: descriptions with SEO, storytelling, and technical structure, no emojis, in the account's language.

Language

The AI writes in the account owner's interface language — Português, Español, or English. Change it using the language selector at the top of the dashboard; future enrichments follow the new choice. Language is not set per supplier or per store.

AI Model

By default, Enrich uses the economy tier (fast, efficient models), at no extra cost. With the AI Model add-on you unlock better models and choose which one to use for each job, in Jobs → New sync → AI Model.

What you get out of it: a stronger model writes more accurate, more persuasive descriptions, which reduces manual editing after the job and helps page conversion. For a store processing a few hundred products a month, standard pays for itself if it saves ~1 hour of text review; premium/ultra make sense when the copy is the main selling point (fashion, cosmetics, gifts).

• Subscribing to a tier unlocks that tier and everything below it (ultra covers premium, standard, and economy).
• If a job requests a model outside your add-on, it's rejected — subscribe to the tier in Billing → AI Model.
• If a model fails during the job, Trinis automatically falls back to the default economy model.

Cache and reprocessing

Enrich uses the same cache described in Configuring Collect (per account + store + EAN, 30 days).

• With "Skip existing products" turned on (the default in New sync), products that already have a cached enriched description are not resent to the AI and do not consume quota.
• Only products that actually go through the AI count against the plan's limit.
• Changed the tone of voice, language, or model and want to reapply it to everything? Turn off "Skip existing products" when creating the job to force re-enrichment.

Best practices

• Be specific in the prompt: target audience, text length range, what to mention (materials, dimensions, care instructions), and what to avoid (emojis, first person).
• Test at small scale: run a job with a low Custom limit, check the descriptions in Products, and adjust the prompt before a full run.
• One tone per account. If different brands need distinct voices, use a per-supplier prompt (contact support) — it takes precedence over the account default.
• A better model where it matters: catalogs where copy influences conversion justify a premium tier; for high volume and standard copy, economy is usually enough.
• The spec sheet helps a lot. Suppliers that expose a spec sheet and a rich description produce better enrichments — that comes from the Collect configuration.

Next steps