Muse Spark 1.3 is one of the large language models that power Muse, Meta's personal agent. And now, with Answer Engine Insights in Profound, you can see how it talks about your brand, which sources it cites, and how it compares with ChatGPT, Gemini, and the other Answer Engines you already track.
However, we want to be transparent that tracking Muse Spark and tracking Muse aren’t the same thing. Muse knows who it’s shopping for, and Muse Spark doesn’t. In our research, giving the model a user’s memory was enough to change which products it recommended. Adding Personas to Muse Spark prompts gives the model some of that context, so you can see how its answers shift for the audiences you care about.
How is Muse different from Muse Spark?
Muse Spark is a large language model that developers can call through Meta's API. Muse is a personal agent that runs on the same family of models, but it's a different product. As we covered in The era of personal agents, Muse pairs the model with rich personal context and defined workflows so it can act on a user's behalf.
Personas are the closest proxy we offer for the personal context Muse uses. They add audience context, such as role, industry, and motivations, to the Muse Spark prompts you choose, and our earlier research found that this kind of context changes which brands Answer Engines recommend. Each persona describes a type of buyer, so it captures part of what Muse knows about an individual user.
What does personal context change?
Personal agents remember a lot about the people they work for, and that memory changes what they recommend. In our personal context study, we asked ChatGPT Instant and Muse Spark the same shopping questions with and without a memory file like the one Muse keeps for each user, across 2,240 answers. Here's what changed.
- Different products. Across both models, repeated answers to the same question usually shared some products. When one answer had relevant memory, and the other didn't, they shared about half as many.
- More personal searches. With relevant memory available, 54% of Muse Spark's search queries included something about the user.
- Shorter shortlists. Muse Spark named 4.5 products per answer with relevant memory, compared with 6.5 without it, and cited 0.7 distinct websites per answer instead of 1.4. Fewer slots per answer means a brand needs a clear reason to be picked for a specific person.
- Tailored explanations. When relevant memory was available, Muse Spark referred to it in 83% of answers and pitched the same product differently to different people.
A prompt sent to Muse Spark carries no memory, so its answers resemble the study’s no-memory answers.
Why does the harness matter?
A harness is everything a product wraps around a model, including its instructions, tools, and interface. We've seen it change results even when the model stays the same.
- Claude and Claude Code use the same underlying models. In our analysis of 24,135 responses, Claude Code searched in 13% of responses compared with Claude's 93%, and the two mentioned only 1 in 5 of the same brands.
- Gemini, AI Overviews, and AI Mode share Google's infrastructure, yet brands saw a median 8-point visibility gap between their best and worst Google model.
Muse's shopping workflow is another good example of this phenomenon. It checks the user's shopping profile, then searches merchant sites in a browser and Meta's product catalog in parallel, and its final recommendations draw on both.
What you can do with Muse Spark in Profound
With Muse Spark in Answer Engine Insights, you can:
- See what the model behind Muse knows about your brand, products, and competitors.
- Find which sources it cites when it searches the web on the topics you track.
- Compare its answers with other Answer Engines, using the same prompts and metrics.
- Run prompts with Personas to see how its answers change for each audience segment you care about.
- Catch inaccurate claims the model makes about your brand before they carry into products built on it.
Treat the results as model visibility. Strong visibility on Muse Spark means your content reaches the model, which is a good sign that Muse may recommend you to a particular shopper, but it's not a guarantee.
The same reading applies to web search providers that agents build on, such as Exa, which Profound also tracks. Visibility in a search index shows whether your pages surface. It doesn't show how an agent writes its queries or ranks what comes back.
What it means for marketers
- The fundamentals of AEO still apply. As we found in The era of personal agents, AI still needs to understand your brand and products to recommend them accurately. Accurate, well-structured pages matter for agents just as they do for Answer Engines.
- Make your products visible to Muse. Include them in Meta's product catalog, one of the sources Muse searches when it shops. The same catalog powers shopping on Instagram and Facebook.
- Make your site easy for agents to use. Muse reads pages in a browser through the accessibility tree. Semantic HTML and clear labels help it understand your site, so test that agents can navigate it.
- Explain who each product is for. In our personal context study, Muse Spark recommended fewer options when it had relevant context, which leaves less room in each answer. Make intended uses, compatibility, specifications, and tradeoffs easy to find, so an agent has a reason to pick your product for a specific person.
- Measure the model behind Muse. Add Muse Spark to the Answer Engines you track and run it with Personas to see what the model knows about your brand and how that shifts for your priority audiences.
What's next
Model coverage is a first step. We'll keep researching how personal agents choose what to recommend and how brands can measure what those agents show their users so that marketers can further optimize their growth strategies.
For more of our research on personal agents, watch The Future of Agentic Attention from Zero Click London.
Get started
Muse Spark is available. Talk to your account team to add Muse Spark to your tracked engines. If you are not yet a customer, book a demo to learn more.
