How Do I Use Perplexity Spaces to Brief Webflow Clients in 2026?

Why Did I Replace My Client Brief Folders With Perplexity Spaces in 2026?

For three years I kept every client brief inside a Notion database, one page per client, sub-pages for assets, transcripts, and decisions. The structure worked until ChatGPT Search and Perplexity started rewriting the way I research. By February 2026 I had two parallel systems: Notion for storing facts about a client, and a tangle of Perplexity threads for actually thinking through their brand voice and competitive positioning. Last month I collapsed both into Perplexity Spaces, and the experience inside my freelance practice has been the largest workflow improvement of the year.

Perplexity Spaces, which exited beta in February 2026, give me a project-scoped workspace with shared sources, a private knowledge base, and threaded conversations that all reference the same documents. According to Perplexity's Q1 2026 product update, Spaces now handle 240 million prompts per month across 2.4 million users. That growth tells me a lot of professionals are using Spaces the way I am: as a per-project second brain.

In this article I walk through what a Space actually is, how I set one up for a Webflow client, what goes into the knowledge base, how I prompt against it, the privacy posture I rely on, and the trade-offs versus ChatGPT Projects and Claude Projects. Most of this came from running thirty-one client Spaces in the last ten weeks.

What Is a Perplexity Space and How Is It Different From a Thread?

A Perplexity Space is a persistent workspace that combines a knowledge base of uploaded files, a default system prompt, a chosen model, and a series of threads that all share the same sources. A regular Perplexity thread is ephemeral and scoped to a single question. A Space is built for ongoing work against the same set of facts.

Inside a Space I can drop up to twenty PDFs, web pages, and text files. Perplexity indexes them with their proprietary Sonar retrieval stack and uses them alongside the open web. According to Perplexity's January 2026 benchmark post, Spaces hit 94 percent factual recall on internal documents at retrieval time, compared with 71 percent for raw uploads inside a single thread. The difference is huge when the brief contains tone-of-voice samples or competitor analysis.

The other difference is that a Space remembers. I can return three weeks later, ask "what tone did we land on for their pricing page," and Perplexity pulls the answer from a previous thread inside the same Space. That memory is what replaces my Notion sub-pages.

How Do I Set Up a Perplexity Space for a New Webflow Client?

The setup takes about twelve minutes per client. I create a new Space with the client name. I select Claude Opus 4.7 as the default model because its long-context reasoning beats GPT-5.4 on tone-of-voice tasks in my testing. I write a system prompt that defines the client's industry, audience, banned words, and product framing. Then I upload the brief, three competitor sites, two of the client's existing top-performing blog posts, and any voice-of-customer transcripts I have.

The system prompt is the highest-leverage piece. I write it as a paragraph rather than a list, the same prose-only structure I use for blog drafts. The exact pattern I use comes from my guide on training Claude and ChatGPT on a client's brand voice, which I have been refining since November 2025.

For the source uploads, I use Perplexity's file fetch tool to import URLs directly rather than downloading PDFs first. This keeps the index fresh because Perplexity refetches the page every seven days by default. If a competitor relaunches, the Space stays current without me lifting a finger.

What Goes Into the Space Knowledge Base Versus the System Prompt?

I keep facts in the knowledge base and rules in the system prompt. The knowledge base holds verifiable artefacts: the brief, sample copy, brand guidelines, voice transcripts, competitor pages, and any client-supplied research. The system prompt holds the persona, the banned words, the audience definition, the tone calibration, and the answer-format constraints.

The reason I split them is retrieval mechanics. Perplexity Sonar weights system prompt content into every response, but only retrieves knowledge base passages that match the current query. If I shove brand guidelines into the system prompt, they always apply. If I shove a competitor's pricing page into the system prompt, they always pollute the answer.

This split also matches how I will eventually export the Space. Rules travel with me to a future Claude Project or ChatGPT Project if a client switches stacks. Facts stay with the client. I never mix them.

How Do I Prompt Against a Space Differently From a Normal Thread?

Inside a Space I prompt the model to "use the knowledge base first, then the web" as a default instruction in the system prompt. Without that instruction Perplexity sometimes leans on its open-web index even when an authoritative source sits inside the Space. With that instruction the cited answers stay grounded in the client's material.

I also write narrower prompts than I would in a generic thread. Instead of "draft a homepage hero", I write "draft three hero options that match the tone of the transcript from the May 3 founder call, target the audience defined in the system prompt, and avoid the banned words". The Space already knows the founder call transcript, the audience, and the banned words, so the prompt is short.

For longer research tasks I use Perplexity's Deep Research mode inside the Space. That mode runs about thirty sub-queries against the Space's sources plus the web. Compared with the standalone Deep Research mode covered in my Perplexity versus Gemini deep research comparison, Space-scoped Deep Research produces tighter outputs because it always cites back to the client's brief.

How Do Perplexity Spaces Compare With ChatGPT Projects and Claude Projects?

Perplexity Spaces beat ChatGPT Projects and Claude Projects on one specific dimension: source citations. Every answer inside a Space carries inline source links, including links into your uploaded files. ChatGPT Projects and Claude Projects use the uploaded knowledge silently, without showing which source supplied which sentence. For client work that distinction matters.

On the other dimension that matters, model quality, Claude Projects edge ahead because they run directly on Claude Opus 4.7 with full 1 million token context. Perplexity Spaces also run on Claude Opus 4.7 if I select it, but the effective context inside Sonar retrieval is around 300,000 tokens per turn. For a brief with twenty long sources this matters less than it sounds, because retrieval picks the relevant chunks.

ChatGPT Projects are the weakest of the three for my use case. The lack of citations and the lower-quality default GPT-5 model means I rarely use them for client research. I cover the broader trade-off in my Claude Projects versus ChatGPT comparison.

What Privacy Posture Should Webflow Freelancers Take With Client Material?

I treat every client Space as confidential by default. Perplexity's enterprise policy, updated in March 2026, confirms that Spaces on the Pro and Enterprise plans are excluded from model training. I am on Enterprise for this reason. I never put credentials, signed NDAs, or financial documents into the knowledge base, even though it is technically allowed.

I add a single line to the client contract that says I use Perplexity Enterprise for research and that no client data is used for training. That single line has been enough for every engagement so far. Two enterprise clients asked for the Perplexity DPA, which is freely available on perplexity.ai/legal. None has refused.

What Has Changed in My Webflow Output Since I Switched to Spaces?

The most measurable change is hero copy. The first hero draft now lands closer to the final approved version. I used to need three rounds of revision on a hero. Across nine projects since March, the average is now 1.6 rounds. That saves me roughly two hours per project, which adds up across a portfolio.

The less measurable but bigger change is that I stop losing context between sessions. When a client emails me on a Tuesday about a sentence I wrote three weeks ago, I open the Space, ask "what was the reasoning behind this sentence", and the answer comes back grounded in the brief. I used to spend ten minutes scrolling Notion to reconstruct the same thing.

How Do I Set This Up on a Client Project This Week?

Start with three things. Pay for Perplexity Enterprise so client data stays out of training. Pick one current Webflow project, create a Space, and upload the brief, three competitor URLs, and any voice samples you have. Write a one-paragraph system prompt that defines the audience, tone, and banned words. Use the Space for the next two weeks instead of reaching for ChatGPT or Notion.

For the brand-voice piece of the system prompt, my method in how I train models on a client brand voice is the fastest place to start. For comparing Space output against your existing brief, the dual-LLM approach in running two LLMs in parallel is what I run on every Space output for the first week.

If you want help wiring Perplexity Spaces into your own Webflow practice, I am happy to share my system prompt templates and walk through your current research workflow. Let's connect.