Sector-Coverage Assistants: Custom GPTs & Claude Projects

A sector analyst answers the same questions every quarter. Who is in the coverage universe? What is each name's positioning? What are the right multiples for this sector? What is the house view? Copy-pasting that context into a fresh AI chat every time is wasteful — and it is why most analysts never get past one-off prompting. The leverage comes from a persistent coverage assistant: a Custom GPT or Claude Project that already knows your names, your house view, your style guide, and your comp set, so every conversation starts where the last one left off.

This lesson is about designing that coverage assistant for the specific shape of an equity analyst's work. It is not a tutorial on the click-path of the GPT Builder — that mechanical "how do I create a Custom GPT" walkthrough is covered in depth elsewhere, and we link to it below so you can build the shell and come straight back here for the analyst-specific configuration.

What You'll Learn

• How to design a persistent coverage assistant around your specific names
• When to reach for a Custom GPT vs a Claude Project for analyst work
• What coverage context to load (and what to keep out)
• How to maintain the assistant as your coverage and house view evolve

Build the Shell First, Then Configure It for Coverage

The plumbing of standing up a Custom GPT (GPT Builder, knowledge files, conversation starters, sharing) or a Claude Project is the same whether you are a marketer or an equity analyst, so we will not re-teach it here. If you have not built one before, work through Build Your First Custom GPT for the fast hands-on version, or ChatGPT Custom GPTs for Business for the deeper team-deployment treatment. Then come back — the rest of this lesson is about what to put inside that shell to make it a competent equity-research associate.

A quick orientation on which tool fits which analyst task:

Custom GPT (ChatGPT): built in the GPT Builder, shareable by link across your ChatGPT Team workspace, supports knowledge files and external actions. Best as the polished, shareable team assistant and for code-interpreter work (running multiples on a CSV).

Claude Project (Anthropic): built inside Claude.ai, shareable across your Claude Team workspace, excellent at long-context reasoning across knowledge files. Best for filing-heavy work — reasoning across a full 10-K plus four transcripts in one session.

For most coverage, build the same assistant in both and use whichever fits the moment.

A Sector Coverage Custom GPT

Imagine you cover US large-cap semiconductors. Here is the GPT setup you would build.

Name: Semiconductor Sector Analyst

Description: Internal AI assistant for our US large-cap semiconductor coverage. Drafts post-print notes, builds comps, and stress-tests theses.

Instructions:

You are the AI assistant for [TEAM NAME]'s US large-cap semiconductor
coverage. Your job is to help draft research notes, build models,
and stress-test theses for the tickers in our coverage universe.

Coverage universe (use these tickers as defaults):
NVDA, AMD, INTC, AVGO, QCOM, MU, AMAT, LRCX, KLAC, MRVL

House view as of [DATE]:
- We are overweight data-center exposure (NVDA, AMD, AVGO)
- Neutral on memory (MU has cycle risk)
- Underweight legacy PC/CPU exposure (INTC)
- Key thesis: AI capex cycle has 18-24 months of runway left

Always:
- Use US GAAP and US dollars unless asked otherwise
- Refer to our covered tickers by ticker, not full name
- For valuation, anchor to EV/NTM revenue for AI exposure and
  EV/NTM EBITDA for mature names
- When you do not know a current number, say so — do not guess

Never:
- Invent customer concentration, capex guidance, or segment numbers
  that I have not provided
- Recommend buying or selling — you draft analysis, not ratings
- Use boilerplate sell-side language like "strong execution" or
  "robust pipeline"

Style: declarative, opinionated where we have a view, hedged where
data is thin. Prefer numbers to adjectives.

Conversation starters:

• "Draft the post-print note for the latest earnings call I'm about to paste"
• "Compare the four tickers I list against the coverage universe on three multiples"
• "Stress-test my thesis on [TICKER] — what would invalidate it?"
• "Build the comp table commentary using the multiples I paste"

Knowledge files:

• Coverage primer (10-page doc explaining the sector, key metrics, KPIs you watch)
• Latest house view memo
• Comp set spreadsheet with peers in coverage and adjacent
• Your firm's research style guide
• Template for post-print notes

That GPT is now a competent associate for your sector. Anyone on the team can use it.

A Claude Project for the Same Purpose

In Claude, the same setup looks slightly different. Create a new Project named "Semiconductor Sector Analyst."

System prompt (Project instructions): Same content as the Custom GPT instructions above.

Project knowledge: Upload the same documents — primer, house view, comp set, style guide, templates.

Default model: Claude's largest context-window model, for filing-heavy work.

You now have two parallel assistants — one in ChatGPT, one in Claude — sharing the same context. The Claude version is better for long-context reasoning (analyzing full 10-Ks); the ChatGPT version is better for code-interpreter tasks (running calculations on a CSV).

What to Put in the Knowledge Files

Curate carefully. Quality matters more than quantity.

Good:

• Your firm's research style guide
• Your sector primer (you wrote this once; the AI uses it forever)
• Latest house view memo
• Comp set spreadsheet
• 3-5 examples of your best prior research notes (for tone matching)
• A glossary of internal abbreviations and house terms

Acceptable but treat with care:

• Historical earnings transcripts (only if recent and clearly dated)
• Industry data (only if you trust the source)

Bad — keep these out:

• MNPI (material non-public information)
• Client account information
• Draft ratings or unpublished price targets
• Anything labeled confidential by your firm
• Notes from one-on-one meetings or private channel checks

If your firm has not set policy on what you can put into GPTs and Projects, ask compliance before uploading anything proprietary. MNPI handling, enterprise-safe plans, and the rest of the data-boundary rules are covered in full in the final Compliance, MNPI & Validating AI Output lesson — treat that lesson as the binding constraint on everything you load here.

Specialized Project Variants

Most analysts benefit from running a few different Projects in parallel:

1. Sector Coverage (described above) — for the day-to-day analysis work
2. Drafting — focused on writing research notes and decks; loaded with style guides and templates
3. Modeling — focused on Excel formulas and DCF assumptions; loaded with template models and assumption frameworks
4. Macro — focused on the macro overlay; loaded with the latest Fed minutes, ECB statements, and your house macro view
5. Compliance — loaded with your firm's research disclosure language and forbidden words; used to scrub drafts before publishing

Building five focused Projects beats trying to cram everything into one giant assistant.

Maintaining Your Assistants

Custom GPTs and Projects rot if you do not maintain them. A maintenance ritual:

• Monthly: Update the house view memo and the comp set
• Quarterly: Refresh the coverage universe (new IPOs, dropped names, M&A)
• Twice a year: Re-review the instructions and remove anything that no longer reflects house view
• Annually: Audit the knowledge files for stale documents and replace them

Treat your assistant like a junior team member — onboarding takes time, but the leverage compounds.

A Real Workflow Example

A semiconductor analyst at a long/short fund built the GPT above. Her actual workflow on earnings day:

1. Paste the transcript into the GPT
2. Run "Draft the post-print note" prompt
3. Use code interpreter to compute the new comp multiples and update the comp table
4. Run "Stress-test the thesis" prompt with the new numbers
5. Edit the draft, add the variant view, send to PM

Time from earnings call ending to a PM-ready note: 35 minutes. Pre-AI, the same workflow took her about 3 hours.

Sharing With the Team

Custom GPTs are shareable inside your ChatGPT Team workspace via a link. Claude Projects are shareable inside your Claude Team workspace. Both let you give every analyst on the team access to the same context.

The most successful teams treat the assistant as shared infrastructure — one person owns it, but everyone contributes feedback when the output drifts from house style. Treat it like internal code: it has an owner, a versioning discipline, and a review cadence.

Key Takeaways

• A coverage assistant exists to make context persistent — build the GPT/Project shell once (see the linked build courses) and configure it around your specific names
• Run one assistant per sector plus focused variants for drafting, modeling, macro, and compliance — beats one bloated catch-all
• Curate the knowledge files — quality of coverage context dictates quality of output
• Never put MNPI or client information into a Custom GPT or Project; the compliance lesson is the binding rule
• Maintain the assistant on a monthly and quarterly cadence — coverage assistants rot otherwise