# Get better results from Attio's AI

URL: https://attio.com/help/reference/how-to-guides/get-better-results-from-attio-s-ai
Breadcrumbs: [Help](https://attio.com/help) > [Reference](https://attio.com/help/reference) > [How-to guides](https://attio.com/help/reference/how-to-guides)

Keep AI tasks focused, accurate, and efficient.

> Available on all plans.
>
> Available for **all** workspace members. **Admins** can see credit usage for each member and workflow, and can limit how much of the workspace's credits Ask Attio uses.
>
> The credits included depend on [your Attio plan](https://attio.com/pricing).

You'll get the best results from Attio’s AI features when your setup matches the task you're doing. Two tasks that look alike can cost very different amounts, and the difference usually comes from how much the AI has to read. This guide shows you how to keep each task focused, choose a model on purpose, carry your work into a fresh chat when one gets long, and check what a task cost.

## Quick tips

These tips are in order of impact. Each one is covered in more detail below.

- Give each task its own chat. Every message re-reads the whole chat, so a long chat uses more credits with each reply. When a chat gets long, carry the key points into a new one with a handoff prompt.
- Tell the AI exactly what to read. In Ask Attio, @mention the records you mean and give a time range, so it doesn't read every matching email or call.
- Choose your model before you start. Switching models partway through a chat makes the next message use more credits.
- Check what a chat cost. Click the ⋮ icon in a chat to see the credits it has used.
- Create saved prompts for tasks you repeat. This way, you can test different models on an identical prompt in fresh chats and compare the output and the credit usage.
- Test AI workflows on one record first. Check the output and the credit usage before you run the workflow on every record.

## Understand what shapes an AI task's cost and quality

Every time you send a message in Ask Attio, the AI re-reads the entire chat before it replies. A chat uses more credits per message as it gets longer, so your twentieth message in a chat uses far more credits than the first one, even when you ask something short.

Three things make a message use more credits:

- **A long chat:** Each message carries everything that came before it.
- **A long break between messages or a model switch:** When you reply within a few minutes, much of the chat is processed at a lower rate. After a break, or when you switch to a different model during a chat, the next message processes the whole chat at the full rate. Reopening a long chat days later uses more credits than starting a new chat with the same request.
- **Large lookups:** A broad question can lead Ask Attio to read many emails or call recordings in one turn, which can make a single message use many times more credits than usual.

It can feel efficient to keep using one chat because it already knows the background. In practice, you pay for that background on every message. Long chats can also produce weaker answers. AI models get less accurate as the amount they have to read grows, so details from earlier in a chat are more likely to be missed.

The model you choose affects cost too, but it's usually a smaller factor than the three points above. The following sections cover what to do about each of the three factors we just mentioned.

## Give each task its own chat, and hand off when one gets long

By default, start a new chat for each new task or when you come back to a chat after a long break. Stay in the same chat for quick follow-ups within a few minutes, while the conversation is still fresh.

A new chat doesn't know what you worked out in the old one, but a handoff solves that. Ask the long chat to summarize what the new chat needs, then paste the summary into a new chat:

- Send the handoff prompt below in the Ask Attio chat you want to continue.
- Copy the handoff Ask Attio writes.
- Start a new chat and paste the handoff to continue the task.

```
I'm continuing this work in a new chat. Write a handoff I can paste into it.

First work out what kind of task this is and where it stands, then choose the structure that suits it best.

Include only what the new chat can't easily get back:
- the goal and where things stand now
- decisions made, plus my preferences and corrections
- approaches tried and rejected, each with a one-line reason, if the new chat might retry them
- what's left to do, ending with the very next step

Leave out anything the new chat can look up, and don't recount how we got here. Don't copy the contents of records, lists, people, calls, or emails.

Don't use @mentions, links, or embeds, because they can't be pasted. Instead, end with a plain-text "References" section that lists every resource the next steps need. Give full, unshortened IDs (I'm explicitly asking for them) and one line on why each resource matters. Use this format:
- Call recordings: title, date, meeting ID, call recording ID (both are needed), and mm:ss timestamps for key moments.
- Meetings: title, date, meeting ID.
- Records: object name, record name, record ID (the object is needed to retrieve it). For people, add their email address.
- List entries: list name, record name, entry ID.
- Emails: subject, sender, date, email ID.
- Notes and tasks: these can't be fetched by ID, so give the title, date, and parent record (object, name, record ID) or meeting ID.

Open the handoff by telling the new chat to retrieve each reference by its ID rather than searching by name, and to find notes and tasks through their parent record or meeting.

Make the handoff as short as possible while still letting the new chat carry on without me repeating anything. Write it as instructions to the new chat, ready to paste.
```

The new chat should pick up the task without you explaining it again. You can [create a saved prompt](https://attio.com/help/reference/how-to-guides/get-better-results-from-attio-s-ai#run-repeated-tasks-as-saved-prompts) for this handoff and add it to any chat by typing `/`.

For details you'd otherwise repeat in every chat, such as your role or preferred output format, add them to your [personal context](https://attio.com/help/reference/attio-ai/ask-attio/manage-ask-attio-settings) instead. Ask Attio applies your personal context to every new chat.

## Point Ask Attio at exactly what it needs

Ask Attio searches calls, notes, emails, and records to answer a question, and it can also perform web research. The broader the question, the more it reads, and the more credits the answer uses.

For example, asking "What's been happening with GreenLeaf?" before a renewal call could lead Ask Attio to read every email and call with the company going back years. Name the record and the time range instead:

```
Summarize our calls and emails with @GreenLeaf from the last 30 days and list any open questions to raise on the renewal call.
```

By default, @mention the objects, lists, users, teams, or records you're referencing and say which period to look at. Ask more broadly when you don't know where the answer is, such as when you're exploring an account you've never worked on. A broad question reads more and uses more credits, so give it its own chat.

## Choose your model before you start

Ask Attio uses **Auto** by default, which sends every request to the same default model. To choose a different model, click **Auto**.

Keep Auto as your default and choose a different model at the start of a chat when the task calls for it. Use this table of examples to pick a model for your task:

| Task                                           | Examples                                                                                                                                      | Start with       |
| ---------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------- | ---------------- |
| Simple: look up, extract, or reformat          | Find a value on a record, pull a date or amount from an email, sort leads into fixed categories, shorten or rewrite text                      | A faster model   |
| Everyday: summarize or draft                   | Prep for a meeting with one account, summarize the last 30 days of calls and emails, draft a follow-up, summarize the week's pipeline         | **Auto**         |
| Complex: judgment across many steps or sources | Build or debug a workflow, find patterns across many deals, plan an account from conflicting signals, qualify leads against detailed criteria | A stronger model |

When a task needs a stronger model, use one, especially for tasks that depend on judgment such as building a workflow or qualifying leads against detailed criteria.

If a model's output falls short, start a new chat with a stronger model and bring the context over [with a handoff](https://attio.com/help/reference/how-to-guides/get-better-results-from-attio-s-ai#give-each-task-its-own-chat,-and-hand-off-when-one-gets-long).

The tradeoff is quality against credits per message. Stronger models cost more each time, and faster models can miss nuance on harder tasks.

Avoid switching models partway through a chat. A switch sends the whole chat to the new model at the full rate, so the next message costs much more, especially in a long chat. To use your preferred model in every new chat, set a default model in your settings. Learn [how to manage Ask Attio settings](https://attio.com/help/reference/attio-ai/ask-attio/manage-ask-attio-settings), and see [Chat with Ask Attio](https://attio.com/help/reference/attio-ai/ask-attio/chat-with-ask-attio) for the models available.

For details on each model, see the vendor documentation: [Anthropic: Sonnet and Opus](https://platform.claude.com/docs/en/about-claude/models/overview), [Google: Gemini Flash and Gemini Pro](https://ai.google.dev/gemini-api/docs/models), and [OpenAI: GPT](https://platform.openai.com/docs/models).

## Check what a chat costs

Each Ask Attio chat shows the credits it has used. In a chat, click the ⋮ icon. **Chat usage** shows the credits the chat has used so far.

![](https://a.storyblok.com/f/234930/648x672/299e913969/cleanshot_2026-10-30_at_09-23-042x.png)

For workflows, the run viewer shows the credits each step used.

![](https://a.storyblok.com/f/234930/848x574/538a70dd9c/cleanshot_2026-10-30_at_09-25-342x.png)

To see the credits you've used across all your chats, navigate to **Account settings**. Under **Personal**, select **Ask Attio**.

Admins can see each member's Ask Attio usage in **Workspace settings** under **Ask Attio usage**, and credits by workflow in **Billing** under **Usage**. Learn more about [managing workspace and seat credits](https://attio.com/help/reference/workspace-settings-billing/manage-workspace-and-seat-credits).

## Run repeated tasks as saved prompts

If you run the same kind of request regularly, save it as a prompt and run it in a new chat each time. Every run starts from a clean slate.

Ask Attio uses whichever model is selected in the chat when you run a saved prompt. Unless the task is [especially high or low in complexity](https://attio.com/help/reference/how-to-guides/get-better-results-from-attio-s-ai#choose-your-model-before-you-start), run the saved prompt on **Auto**.

As a test, you can also load the saved prompt in a fresh chat, choose a model from the **Auto** menu before you send it, and compare the results with the output from a stronger model. If the output is good enough with the faster model, use that model for that prompt. A faster model can miss edge cases, so check its output.

Learn [how to create and use Ask Attio prompts](https://attio.com/help/reference/attio-ai/ask-attio/create-and-use-ask-attio-prompts).

## Test and tune AI blocks in workflows

AI blocks in workflows such as the [Custom agent](https://attio.com/help/reference/automations/workflows/workflows-block-library) and [Web agent](https://attio.com/help/reference/automations/workflows/workflows-block-library) blocks are charged by how much the AI reads and writes on each run, so their cost varies. A small difference per run adds up quickly if your workflow is triggered a lot.

Once a workflow runs at scale, choose a specific model for the Custom agent blocks within it. Use the model that gives the right output on a few sample records. For simple, high-volume tasks that's often a faster model. Pick a stronger model when accuracy matters more than cost, such as qualifying high-value leads.

To test a workflow:

- Review what each block does.
- Run the workflow once.
- Open the run viewer and check the output and the credits each step used.
- Run it on a small batch before you run the workflow at scale.

You should see the output you expect and a cost per run you're comfortable multiplying by your number of records.

AI attributes cost a fixed number of credits per record. Learn the [best practices for AI attributes](https://attio.com/help/reference/attio-ai/ai-attributes).

## Related resources

- To set your default model, add personal context, or have an admin add workspace context that applies to every chat, see [Manage Ask Attio settings](https://attio.com/help/reference/attio-ai/ask-attio/manage-ask-attio-settings).
- For tasks you repeat, [create and use Ask Attio prompts](https://attio.com/help/reference/attio-ai/ask-attio/create-and-use-ask-attio-prompts) so you write the prompt once and run it in a fresh chat each time.
- If you're planning how your workspace spends credits, [Manage workspace and seat credits](https://attio.com/help/reference/workspace-settings-billing/manage-workspace-and-seat-credits) explains what each AI feature costs and how admins can cap Ask Attio's share.
- When you build an AI workflow, the [Workflows block library](https://attio.com/help/reference/automations/workflows/workflows-block-library) covers the inputs and settings for the Custom agent and Web agent blocks.
- Before you run an AI attribute across a full view, read the [best practices for AI attributes](https://attio.com/help/reference/attio-ai/ai-attributes).
- To see how Attio and its model providers handle your data, read the [Attio AI Policy](https://attio.com/help/reference/attio-ai/attio-ai-policy).
