Skip to content
Standard Deviation

StandardGPT

StandardGPT is the conversational interface layered on top of the three modules. Chat with your brand — but the model has read your StandardGraph, your campaigns in StandardPlanning, and your assets in StandardCreative, so it answers from what’s actually there.

What it can do

  • Answer questions about the brand graph. “What does our voice profile say about formality?” “Which competitors did we add last quarter?” “What’s our top AI-search category?”
  • Summarize campaigns. “Walk me through the Q3 launch campaign.” “What was the strategic rationale on the SKU launch?”
  • Reason across modules. “Why did we move away from Audience B last quarter?” “Which creative assets from the spring launch scored highest on Brand Recall?”
  • Draft net-new content. Headlines, captions, briefs, descriptions — but grounded against the loaded brand kit, same as StandardCreative generations.
  • Run open-ended explorations. “What positioning gap could a challenger brand exploit against our top three competitors?” “If our category’s effective frequency is 8, what could shift it to 6?”

What it does not do

  • It doesn’t modify the brand graph. StandardGPT is read-only against StandardGraph. To update the brand profile, edit it directly in StandardGraph.
  • It doesn’t place media. Suggestions are advisory; actual buying happens downstream.
  • It doesn’t hallucinate access. If a brand graph is incomplete, GPT will say so rather than fabricate a profile. If a campaign doesn’t exist in StandardPlanning, GPT won’t pretend to know what was in it.

How retrieval works

Each conversation thread is scoped to one brand. When you ask a question, StandardGPT:

  1. Embeds your message + a window of recent conversation context
  2. Retrieves the top-K relevant fragments from across the brand’s graph, planning, and creative records
  3. Injects retrievals + brand-kit metadata into the model call (Gemini 2.5 Flash by default)
  4. Streams the response back

The retrieval index is refreshed on graph + campaign + asset writes, so answers stay current as the brand evolves.

Thread history

Conversations are stored per-brand under /standard-graph/standard-gpt-history. Useful for:

  • Returning to a previous answer
  • Handing a thread to a teammate
  • Auditing what was asked + answered when the brand made a decision

Threads can be starred for quick re-access; bulk-delete is also supported.

How clients use StandardGPT in practice

The most common pattern: a marketing lead opens StandardOS first thing in the morning, asks StandardGPT a 30-second question (“what changed in the competitive landscape this week?”), gets an answer with citations into Social Listening and Search Visibility, and either acts on it or routes the insight to the team. Five-minute habit, compounds over time.

A second pattern: mid-campaign, asking StandardGPT to draft something specific against the loaded campaign context. “Write me 5 headline variants for the launch creative we’re working on” — drafts come back grounded on the campaign strategy + locked audience, not generic.

What StandardGPT is not

  • Not a general-purpose LLM front-end. If you want to chat about the news or write a personal email, use ChatGPT. StandardGPT is intentionally brand-bounded.
  • Not a customer-facing chatbot. Don’t embed it on your website. It reads your internal brand kit; that surface should stay private.

Where to go next