Summarize record

The Summarize record block (AI step) uses AI to write a text summary of a record from its attributes.

The Summarize record block (AI step) uses AI to write a text summary of a record from its attributes. It reads the values stored on a record and returns a short write-up. The block returns that summary as a run variable only. It never writes the text back to the record on its own. To store the summary, pair this block with Update record.

Category	AI step
Credit cost	Variable (token-based)
Engine	New Workflows engine (June 2026+)
Plan availability	All paid plans
Limits	None documented.

Inputs

Input	Type	Required	Default	Notes
Record	Record	Yes	None	The record to summarize. The block reads its attributes as the source material.
Instructions	Text	Yes	None	What to include, omit, and emphasize, plus the target length.

Outputs

Output	Type	Example	Notes
Record summary	Text	Acme Corp is a 200-person logistics SaaS in renewal at $48k ARR; primary contact Priya Sharma, last touch 14 days ago.	The generated summary. Nothing is persisted until a later block writes it.

Reference the generated text downstream as {{ steps.summarize_record.record_summary }}.

Behavior

The block sends the record’s attributes and your instructions to the model. It then returns the model’s text response as the record_summary output. The block reads from the record but does not write to it. So the summary exists only as a run variable until a later block saves it.

Token cost scales with how much the block sends and receives. A token is a small chunk of text, and you pay for the tokens that go in and the tokens that come back. A record with long notes and many populated attributes costs more than a sparse one. A request for three paragraphs costs more than a request for one sentence.

The block is non-deterministic. That means the same record and instructions can produce slightly different wording on each run. Output length and tone depend almost entirely on how precise your instructions are. Vague instructions drift in length and emphasis from run to run.

Gotchas

• The summary is never saved automatically. The block only emits a run variable, so the text disappears when the run ends unless something consumes it. First, create a long-text attribute to hold the summary. Then follow Summarize record with Update record and map {{ steps.summarize_record.record_summary }} into it.
• Summaries drift in length and focus. When instructions leave length, audience, or source priority unstated, the model picks them itself. It then changes its mind between runs. State the target length, the reader, and which attributes matter. For example: “Two sentences for an account executive; lead with renewal date and open opportunities”.
• The model can only summarize what the record holds. The block sees attribute values, not external context. It cannot invent figures the record lacks, and it may pad with generic phrasing when fields are empty. Enrich or populate the record before summarizing. (Enrich means filling in missing data, such as company size or industry, from another source.) You can also tell the model to omit anything it cannot ground in an attribute.

Examples

Drafting a sales brief before a call

A revenue operations team wants a one-line brief on each company so account executives walk into calls prepared. The workflow runs from a Record command on the company. A Record command lets a user start the workflow from a button on the record. The workflow summarizes the record, then stores the result on a Sales brief long-text attribute.

Field	Value
Record	{{ trigger.record }}
Instructions	Write a two-sentence brief for an account executive. Lead with renewal date and ARR, then the primary contact and last touch.

The block returns Acme Corp renews 2026-09-30 at $48k ARR; one open expansion opportunity. Primary contact Priya Sharma at [e-mail], last touched 14 days ago.. A following Update record block then writes that text to Sales brief.

Producing a structured deal recap

A team wants a fixed-shape recap on every deal so the format stays consistent across reps. Giving the model an explicit template constrains the output. That keeps the summary short and scannable.

Summarize this deal in exactly three labeled lines, each under 15 words:
Stage: <current stage and days in stage>
Risk: <one risk, grounded in an attribute>
Next: <the single next action>
Omit any line you cannot ground in a record attribute.

The block returns a three-line recap such as Stage: Negotiation, 11 days. Risk: No economic buyer identified. Next: Confirm signer before Friday.. A later Update record block can then store it.

Credit cost worked example

Summarize record is token-priced. Each run costs in proportion to the record’s size and the requested summary length. Suppose a team runs it on 400 new companies a month with short, well-scoped instructions. Each run reads a modest set of attributes and returns two sentences. So the per-run token cost is small but not zero. Across 400 runs it adds up against the workspace’s monthly credit allowance.

The price is token-based, not a flat per-run charge. So the lever you control is prompt size. Trimming the instructions and limiting which attributes the model reads lowers the cost of every run. For the exact token-to-credit conversion and your plan’s monthly allowance, see Credits & pricing explained.

Source

Attio Workflows block library.

Last reviewed against the Attio Help Center, June 2026.

See also

• Classify record
• Classify text
• Update record
• Variables & data passing
• AI ICP-fit research and routing