What This Update Actually Is
Before this update, HubSpot's Agents and Assistants could only read the text inside your knowledge sources. Any image sitting in a Knowledge Article, Blog Post, Landing Page, or Site Page was invisible to them. They skipped right over it.
Now, the Knowledge Vault automatically extracts up to 10 inline body images per page. It downloads each image, uses a large language model to generate a semantic description of it, and indexes that description alongside your text. Agents and Assistants then pull those descriptions into their context window at query time.
There's one important boundary to know: only body-content images are extracted. Header images, footer graphics, and decorative elements are intentionally excluded. That's a smart call. Decorative images add noise without adding meaning.
Existing content is already being re-processed through a migration job. You don't need to republish anything. Citations in Agent responses point to the parent page, not the image file directly.
Why HubSpot Shipped This
Think about the kind of content that lives in a knowledge base. Product spec sheets. Installation guides. Troubleshooting flows. Step-by-step how-to articles. Screenshots with callouts. Almost all of it has images, and almost all of those images carry information that text alone can't replicate.
When your AI-powered Support Agent couldn't see those images, it gave incomplete answers. A human reading the same article would say, "Look at the diagram on step three." Your Agent had no idea that diagram existed. That gap quietly eroded trust in AI support.
HubSpot already handled this for PDFs and Office documents. This update brings HTML content up to the same standard. The goal is simple: what you author should be fully available to the Agents working on your behalf.
How to Use It Step by Step
The short version is: you don't have to do much. But there are things you can do to get more value out of this change right now.
- Audit your highest-traffic knowledge articles for image quality. Blurry screenshots, generic stock images, or images without meaningful context won't generate useful semantic descriptions. Replace them with clear, specific visuals that carry real information.
- Add descriptive alt text to every body image. The LLM uses the image itself to generate a description, but alt text gives it additional context. Write alt text as if you're explaining the image to someone who can't see it.
- Republish or update any article that has stale or low-quality images. Publishing a fresh version triggers the automatic extraction process immediately rather than waiting for the migration job to reach it.
- Test your Agent responses. Pull up your Breeze Agent, ask it questions that previously required a human to reference a diagram or screenshot, and see how the responses have changed. Look for specificity and accuracy.
- Watch your image budget per page. Only 10 inline body images are extracted per page. If a page has more than 10, prioritize the most information-dense ones by ordering them near the top of the body content.
What It Touches in Your HubSpot Strategy
This update ripples further than it first appears. Here's where it lands across your portal.
Service Hub: Your Breeze Customer Agent now has a fuller picture of every support article. For companies with detailed product documentation, this means fewer escalations from AI to human reps because the Agent can actually walk someone through a visual process.
Content Hub: Blog Posts, Landing Pages, and Site Pages are now part of the knowledge surface that Agents can draw from. If you're using a Breeze Assistant to help humans navigate your site content, those Assistants now have access to the full visual story you've built.
Key Takeaway
Every image you publish in a Knowledge Article or site page is now part of your AI's working knowledge. Treat your visuals with the same intentionality you bring to your copy.
Knowledge Vault and Breeze AI infrastructure: This closes the gap between PDF and Office document support (which already handled images) and HTML content. Your Knowledge Vault is now a genuinely multimodal resource. That's a meaningful shift for how you think about content authoring going forward.
This update also connects to the broader pattern of HubSpot expanding what Breeze Agents can perceive and act on. If you've been following how Agents are evolving, you might want to read about
how Breeze Agents now connect to tools like Gong and Amplitude via MCP servers. Together, these updates point toward Agents that see more, know more, and act more accurately.
Key Takeaway
Multimodal AI support isn't a future feature anymore. It's live in your portal today. The humans your Agents serve will get more accurate, more complete answers as a direct result of this change.
One strategic note: this also raises the stakes for content quality. A well-labeled, informative diagram now actively improves your AI's output. A cluttered or meaningless image does nothing and could introduce noise. Your content team's visual hygiene just became an AI performance variable.
Who Should Care Most
Not every team feels this equally. Here's where the impact is sharpest.
- Customer support teams running Breeze Agents on product-heavy knowledge bases: this directly reduces the accuracy gap between your Agent and a trained human rep.
- Marketing ops leaders who manage Content Hub sites: your Landing Pages and Site Pages are now richer knowledge sources for any Assistant built on top of them.
- SaaS and product companies with visual documentation: spec sheets, UI walkthroughs, and annotated screenshots are exactly the kind of content this update was built for.
- Content strategists responsible for knowledge bases: your content architecture decisions now have a direct line to AI response quality. That's both a responsibility and an opportunity.
If your company relies on visually rich how-to content to support humans post-purchase, this update is for you. All hubs and all tiers get it. There's no paywall here.
If you're also using Breeze Assistant for content work, you'll want to see how Breeze Assistant Projects let you group related chats with shared instructions and knowledge. Pairing that workspace feature with richer, image-aware knowledge sources makes for a significantly more capable AI layer.
George's Take
I've been inside hundreds of HubSpot portals, and one pattern comes up constantly: companies spend real time building out their knowledge base, they include screenshots and diagrams because that's what makes the content actually useful, and then they turn on AI support expecting it to just work. It never quite did. The Agent would answer around the image instead of with it. That frustration is real, and this update directly addresses it. What I want every team to take away is this: your images are now first-class knowledge assets. Treat them that way. Write meaningful alt text, use clear visuals with intent, and audit your existing articles for image quality before you assume the AI will fill in the gaps for you.
“Your images are now first-class knowledge assets. The AI can finally see what your customers see. That's not a small thing.”
If you want help auditing your knowledge base, setting up Breeze Agents the right way, or building a content strategy that actually makes your AI smarter, let's talk. Book a strategy call with the Sidekick team and let's look at your portal together.
Frequently Asked Questions
Which HubSpot content types now support inline image extraction for AI Agents?
HubSpot now extracts inline images from Knowledge Articles, Blog Posts, Landing Pages, and Site Pages. Only body-content images are included. Header, footer, and purely decorative images are excluded. Up to 10 images per page are extracted and semantically indexed for use by Agents and Assistants.
Do I need to do anything to enable inline image support for my HubSpot Agents?
No action is required. HubSpot automatically re-processes existing content in your Knowledge Vault through a background migration job. New or updated pages are processed at the time of publishing. The feature is available across all hubs and tiers with no configuration needed.
How does HubSpot describe images for AI Agents if there's no text alternative?
HubSpot downloads each qualifying inline image and uses a large language model to generate a semantic description of the image content. That description is indexed alongside the page text. Alt text you've added to images provides additional context and can improve the accuracy of the generated description.
What happens when a knowledge page has more than 10 inline images?
Only the first 10 qualifying body-content images are extracted per page. If your article contains more than 10 images, prioritize placing your most information-dense visuals earlier in the page body. Decorative and layout images are excluded from the count entirely.
Will AI Agent citations point directly to the image or to the parent page?
Citations in Agent and Assistant responses point to the parent document, not the individual image file. So if an Agent references information from an image in a Knowledge Article, it will cite the article itself. This keeps the citation experience clean and consistent for the humans receiving support.
How does this update compare to HubSpot's existing image support for PDFs and Office files?
HubSpot already extracted and indexed images from uploaded PDFs and Office documents. This update brings HTML-based content, specifically Knowledge Articles, Blog Posts, Landing Pages, and Site Pages, up to the same standard. All major content types in your Knowledge Vault now benefit from full visual context for AI Agents.
