The AI Condition Rules node sorts contacts into branches using AI. You describe each branch in plain language, and AI routes every contact to the best match. Contacts that don't clearly match any branch follow the Alternative branch.
Use it when regular Condition Rules filters are not enough: AI understands meaning, not just exact field values. You'll find it in the Rules group, next to the regular Condition Rules node.
What data the AI sees
For each contact, AI receives these fields (as of today):
Full name
Position (current job title)
Headline (LinkedIn headline)
About (LinkedIn summary)
Company name
Location
Profile language
Not included yet: contact's posts and activity, company details (industry, size, description), email data, custom fields. Keep this in mind when writing branch descriptions — AI can only use what it sees.
The node classifies each contact from this profile data only — not from your conversation or message history with them. Only non-empty fields are sent to the AI.
We keep extending the set of fields the AI receives — new fields are added regularly, and this article always reflects the current list. If a field you need is missing, let us know via chat.
How to reference contact fields
There is no variable syntax here — just mention fields in plain text. AI understands references like:
If the position mentions both ownership and management, prefer the Owners branch.
Judge by the company name, not the person's job title.
Do not use {{ variables }} — they won't be rendered in this node.
Credits
Classification runs on the built-in AI model and uses your team's AI credits from the subscription: 1 credit per 1,000 tokens, typically 1–2 credits per contact (at least 1 credit per contact).
If your team runs out of AI credits for the month, classification stops before it runs: the contact isn't routed, its task fails with a "You have reached your LLM credits limit" error, and processing resumes once you top up or your plan renews.
Which AI model is used
Classification always runs on GetSales' built-in AI model. You can't choose the model for this node yet — the node form only contains branches and Classification instructions, with no model selector. Choosing your own model here is planned for a future update.
Custom LLM keys added under Integrations are not used by this node — it always runs on the built-in model, regardless of any custom LLM you've connected elsewhere.
Setting up the node
Branches — add at least 2 branches (and up to several). The branch description is the matching criteria: write it as "which contacts belong here", with examples. The clearer the description, the better the routing.
Classification instructions (optional) — extra rules for the AI that work on top of branch descriptions. Use them to say which field matters most ("judge by position, ignore the headline") and what to do when a contact fits two branches at once ("if someone is both an owner and a marketer, put them in Owners"). Note: if a contact matches several branches and you haven't given such a rule, AI picks the branch it considers the closest — branch order does not set priority.
Alternative branch — always the last branch, and it needs no description. This is where contacts go if nothing matches clearly (or if the AI returns nothing / an invalid result). Contacts with empty profiles also land here.
Reviewing why a contact was routed
For every contact, the node saves the AI's decision — the branch it picked, a short reason, and the keywords it based the decision on. Open the contact's task to see exactly why it was routed the way it was; this makes it easy to debug and refine your branch descriptions.
Popular use cases
1. Route by role / seniority
Classification instructions:
Base the decision on the position first, then the headline. If a person holds
multiple roles, ownership roles take priority.
Branch #1 — Decision makers:
Business owners and top-level executives: founder, CEO, president, owner,
managing director, C-level, VP and above.
Branch #2 — Sales & Marketing:
Sales, marketing, growth and business development roles of any seniority —
unless the person is also a company owner or C-level executive.
Why AI instead of a regular filter? A position filter matches exact words — it misses "Co-Founder & CMO", non-English titles ("Geschäftsführer", "gérant", "директор") and everything that lives only in the headline. AI reads both fields and understands synonyms in any language.
2. Route by company type
Classification instructions:
Classify by the TYPE OF COMPANY, not the person's job title. A CEO of an agency
and an intern at the same agency belong to the same branch.
Branch #1 — Tech & SaaS:
People working at software product companies: SaaS platforms, developer tools,
fintech. The company builds and sells its own software.
Branch #2 — Agencies & Services:
Companies selling services to external clients: marketing agencies,
IT outsourcing, consulting, recruiting.
This is not possible with regular filters at all — there is no "company type" field.
3. Route by region
Classification instructions:
(can be left empty for simple cases like this)
Branch #1 — Europe:
leads from Europe
Branch #2 — North America:
leads from the USA and Canada
Branch #3 — Middle East:
leads from gulf countries — Saudi Arabia, UAE, Qatar, etc.
You can do this with a location filter — but you'd have to list every country (and spelling variant), and any country you forget sends the contact down the wrong branch. AI knows "Europe" includes Lithuania and "San Francisco Bay Area" is the USA. In our tests, contacts from the described regions were routed with 98%+ accuracy even with one-line branch descriptions.
4. Route by business model: B2B vs B2C
Classification instructions:
Classify by the COMPANY's business model, inferred from the company name and
the person's headline and about. Ignore the person's role.
Branch #1 — B2B:
The company sells to other businesses: enterprise software, professional
services, wholesale, industrial products, B2B SaaS.
Branch #2 — B2C:
The company sells to consumers: e-commerce brands, retail, apps for consumers,
restaurants, personal services.
There is no "business model" field anywhere — but AI reliably infers it from context ("helping brands sell on Shopify" → B2C e-commerce, "enterprise data pipelines" → B2B). A must-have when your offer only makes sense for one of the two.
Best practices
Describe exclusions explicitly. If Customer Success should NOT count as Sales — say so: "…unless the role is customer success or support". AI tends to pick the closest branch rather than Alternative, so draw the borders yourself.
One axis per node. Don't mix "by role" and "by region" in one node — chain two AI Condition Rules nodes instead.
Give examples in descriptions. "e.g. EPAM, Cognizant" or "e.g. 'gérant', 'chef d'entreprise'" noticeably improves matching on ambiguous cases.
Test on a small batch first. Add 10–20 contacts, check the split, adjust descriptions, then launch the full list. Re-entering contacts re-classifies them with the updated descriptions.
Expect gradual processing on large lists. Classification runs asynchronously in batches; a few hundred contacts take tens of minutes. This does not block the rest of your automation.
Contacts with empty profiles (no position, no headline) go to the Alternative branch — plan that path in your flow.