Context management in Spark

Ben Allen

Updated June 05, 2026 20:15

Note: Productboard Spark is currently in open beta and available to all customers. The beta is a standalone experience that doesn't integrate with existing Productboard workspaces.

If you're on an Enterprise plan and workspace integration is important to you, contact your Productboard representative to discuss options. Otherwise, you can join the Spark beta here.

The difference between a generic AI response and one that fits your product reality comes down to context. When Spark understands your personas, strategy, and past decisions, it delivers outputs grounded in how your team actually works, not generic advice that requires heavy editing.

There's a practical limit to adding context too: AI systems can only process so much information at once. You can't point Spark at every document your company has ever created and expect useful responses. More context isn't always better. Learning to curate what Spark draws on is a skill that pays dividends in output quality.

Spark offers two approaches to context:

Shared context: information you store in a dedicated folder that's accessible across all Spark conversations. Spark will try to use the most relevant documents from this folder automatically.
Chat-specific context: documents you add to specific chats when you need them. Think of this as on-demand context for focused tasks.

This article explains how both approaches work and how to build a shared context that helps your entire product team move faster.

If you're looking for information about adding context from external sources like support documentation or GitHub via integration, see Connect external tools to Productboard Spark

How Spark differs from search-based AI tools

Some AI tools take a search-first approach: they index everything in your organization, then filter for relevance when you ask a question. Tools like Glean work this way.

Spark takes a hybrid approach. It automatically pulls in relevant documents from your shared context (your Product Strategy Context folder), and you decide what additional documents to add to each conversation. This means:

No surprises: Responses can only be influenced by the documents you've chosen to include.
Automatic relevance: Spark selects the most useful shared context for each conversation from your company knowledge, so you don't have to manually reference foundational documents.
Transparency: Spark shows what context it used for any output, so you can verify the sources.

This approach balances convenience with control. Spark retrieves information from the shared context folder automatically, but you curate what's in that folder and what you add to each chat.

Company knowledge vs. chat-specific context

Understanding the distinction between these two approaches helps you organize information effectively.

Company knowledge

Data and documents in the Company knowledge section of your workspace are available to every Spark conversation across your workspace. You don't need to reference these documents directly; Spark selects the most relevant ones (typically around three) based on what you're discussing.

This approach works well for foundational information that applies across initiatives: your product strategy, target personas, company OKRs, and competitive landscape. When this information lives in company knowledge, your entire team builds organizational memory that compounds over time. Everyone's Spark conversations can draw on the same baseline knowledge without anyone needing to re-explain the basics.

When Spark recognizes you're generating a standard document type (like a PRD or product brief), it searches for matching templates, within your Templates folder, to structure the output. You can also prompt Spark to use a specific template.

Chat-specific context

Chat-specific context refers to documents you deliberately add to a particular conversation. This information stays within the chat in which it's invoked.

Chat-specific context is lost when you start a new chat. Use it to reference documents that don't live in the shared context folder.

To add chat-specific context, use the @ Context button below the chat window, or @-mention a document in the chat itself.

If you want to carry results from one conversation into another, ask Spark to summarize the discussion, save that summary as a document, and then include it in your new chat.

Use chat-specific context when you need focused input for a particular task. If you're exploring a specific feature area, you might attach the relevant PRD and recent customer feedback related to that feature. This keeps the conversation targeted without pulling in unrelated information.

What context you can add to specific chats

You can bring many different types of context into individual conversations:

Productboard items, documents, and notes.
Highlighted text from within documents.
Filtered feedback from insights boards.
Files from Google Drive or Confluence.
Live data from connected tools like Notion, Amplitude, or Linear.

Spark can make use of Google Docs, Slides, Sheets, Confluence Pages, plain text, CSV, markdown, and PDFs.

Note: To reference data from external sources, you must set up a connector.

⚠️ Limitation: Spark currently cannot retrieve, create, or update core product entities, such as features, releases, objectives, or initiatives. We're working on more integrated experiences and you can upvote this on our portal.

How to add context to Spark

There are a few ways to provide context in a Spark chat:

Reference items in your workspace

You can reference specific items in your workspace by @-mentioning them or pasting their URL in the chat field, or by adding them to a chat with the Context button.

Example: “Hey Spark, please summarize the delivery risk on @Initiative_X, taking into account the progress, timeframe, health status and associated dependencies.”

Spark can also read entities displayed on boards which have been added to chats as context.

Example: “Which features have no owners on this board?”

The following item types can be used as specific context:

Products	Objectives	Releases
Components	Key results	Companies
Features	Initiatives	Users
Subfeatures	Release groups

Spark can also search for, identify, and analyze items (and relationships between items) through natural language. This is useful when you need to work with several items that match certain criteria, or items you aren't familiar with (as opposed to @-mentioning specific items).

Example: Here are some natural-language queries you could ask Spark about your workspace items:

"What features are in the 17.0 release?"

"Refine the problem statement based on subfeatures and linked objectives."

"Find all features in delivery where effort is more than 10."

Spark decides whether to return results inline in the chat window or render a temporary board based on result size, field count, and entity diversity.

If you want to preserve a temporary board, you can save it the same way you would any other board.

Note: Spark always respects each user's workspace permissions when searching and surfacing information. You won't be shown anything you aren't supposed to see just because you ask Spark nicely.

Reference Productboard documents

Store foundational documents in the Product Strategy Context folder to make them available across all chats. For documents you want to reference in specific conversations, use @-mentions to pull them in as needed.

To learn how to create and manipulate Productboard documents in Spark, see Getting Started: Orientation, context, and templates.

Reference highlighted text from within a document

With a document open, you can highlight text and click Ask Spark in the popup menu to add that text to your active Spark chat. You can then have a focused discussion with Spark about your highlighted text, or ask it to make edits that target only that snippet.

You can add multiple snippets at the same time to focus Spark on several specific areas of the document.

Reference documents and data from external tools

Connectors allow you to integrate your Spark workspace with other tools that house data and documents. Once the right connectors are set up, you can attach files from repositories like Google Drive and Confluence, or use conversational language in your prompts to call data from tools like Notion, Amplitude, or Linear.

To learn how to set up connectors, see Connect external tools to Productboard Spark.

Reference notes and insights boards

You can reference filtered feedback boards as context for your prompts. This connects Spark directly to what your customers are saying.

For example, filter an insights board to show feedback from enterprise customers, then ask Spark to summarize common pain points. The AI draws on actual customer voices rather than generic assumptions.

If your customer feedback lives outside Productboard, you can use integrations to bring it into Spark. For example, if you store feedback in a Google Spreadsheet, attach that file directly to your chat.

For more instructions, see Working with feedback notes.

If you don't have any notes in your workspace yet, check out this video:

Best practices for building shared context

Your Spark workspace will function more efficiently if you set up your system of context according to the following principles:

Start with foundational documents: Begin by adding your most-referenced documents to the Company knowledge section: personas, product strategy, OKRs, and competitive intelligence. These form the baseline that Spark uses across all conversations. Align with your team on what belongs in shared context. Agreement on foundational documents means everyone's Spark conversations draw on the same organizational knowledge.
Establish naming conventions: Clear document titles help both humans and Spark find the right information. Use consistent patterns like "PRD: [Feature Name]" rather than "feature doc v3 final." When documents are well-named, Spark can retrieve relevant context more accurately, and your team spends less time searching.
Maintain document health: Archive outdated documents rather than deleting them as you may need to reference past decisions. Mark documents with status indicators: Draft, In Review, Approved, or Deprecated. Clean context produces better outputs. If Spark draws on outdated information, the responses will reflect that.
Review and refine regularly: Context compounds, which means stale information accumulates too. Review your company knowledge quarterly to archive documents that no longer reflect current strategy. Treat the Company knowledge section like a shared knowledge base. Regular maintenance keeps it valuable.