Build a Startup Review Panel

Create three AI reviewer personas to get multi-angle feedback on your startup idea.

Introduction

30 Second Summary

When founders want honest feedback on a startup idea, they usually have to book separate meetings with investors, advisors, and potential customers. Each person evaluates the idea through a completely different lens, and getting all three perspectives takes time and connections most early-stage founders don't have.

In this project, you will build three AI-powered synthetic personas as custom GPTs in ChatGPT, each grounded in real-world research from Perplexity. You'll feed them a startup idea and walk away with concrete, multi-angle feedback in minutes.

What You'll Build

Picture pasting your startup idea into three different custom GPT reviewers and getting back an investor's market analysis, an advisor's assumption stress-test, and a customer's honest objections, all grounded in real research you gathered yourself.

By the end of this project, you'll have:

A VC Reviewer custom GPT you can send any startup idea to and get back a structured investor evaluation covering market sizing, unit economics, and competitive moats.
A Startup Sparring Partner custom GPT that challenges your assumptions, pokes holes in your logic, and stress-tests your thinking like a seasoned advisor would.
A Customer Interview Simulator custom GPT you can pitch your product to and hear realistic objections, buying hesitations, and unmet needs you hadn't considered.
Secret Mission: Build a Review Panel Synthesis GPT that consolidates all three reviewers' outputs into a single prioritized action plan.

Are there any prerequisites?

You'll need a ChatGPT account with GPT creation access. New GPT creation requires a Business, Enterprise, or Edu workspace. Personal accounts (Free, Go, Plus, and Pro) cannot create new GPTs.

You'll also need a Perplexity account (free or Pro) and a startup idea or product concept to use as your test input throughout the project.

Before We Start

Before we dive in, let's lock in what you're about to build and why it matters to you.

Set Up Your Tools and Gather Materials

Two platforms power this project. ChatGPT is where you'll build custom GPTs as your reviewer personas. Perplexity is where you'll research the real-world frameworks that make those personas effective.

Before diving into persona creation, you need to confirm you have access to the right features on each platform. Without verifying now, you could hit a wall mid-project when you try to create your first GPT.

In this step, get ready to:

Verify your ChatGPT account supports creating new GPTs.
Confirm Perplexity Research mode is available in your account.
Gather your startup materials for consistent testing across all three reviewers.

Verify your ChatGPT GPT creation access

Custom GPTs are reusable AI assistants you configure with their own name, instructions, and reference materials. You'll create three of them in this project, so you need an account that supports GPT creation.

Open your browser and visit https://chatgpt.com/gpts.
Sign in to your ChatGPT account if prompted.
Look for the Create button on the page.

✔️ I see the Create button

Your account supports GPT creation. You're all set for this project.

ⓧ I don't see the Create button

GPT creation requires a Business, Enterprise, or Edu workspace. Personal ChatGPT accounts (including Free, Go, Plus, and Pro) cannot create new GPTs.

If your account doesn't support GPT creation, you have two options:

Switch to a ChatGPT workspace that has GPT creation enabled. Ask your workspace admin if GPT creation permissions are turned on.
Use ChatGPT Projects as a fallback. Projects let you store persona instructions and reference materials in a persistent context, even though they don't create a standalone, shareable assistant. You can follow along with the project by creating a Project for each persona instead of a custom GPT.

What is a custom GPT?

A custom GPT is a version of ChatGPT configured for a specific purpose. It combines its own instructions, knowledge files, and selected capabilities into a tailored assistant.

Each GPT starts every conversation fresh. It follows only the instructions and knowledge you've set up, with no memory of previous chats.

💡 Use a web browser for GPT building

Building and editing GPTs is limited to the web experience. Mobile apps support using GPTs but not building them. Make sure you're working in a desktop or laptop browser throughout this project.

Confirm Perplexity Research mode is available

Research mode is Perplexity's advanced research feature. It performs dozens of searches, reads hundreds of sources, and synthesizes findings into a comprehensive report with citations. You'll use it to ground each persona in real-world expertise.

Open perplexity.ai in your browser.
Sign in to your Perplexity account.
Click the mode selector in the search box.
Confirm that Research appears as an option in the dropdown.

✔️ I see Research in the mode selector

Research mode is available. You're ready to go.

How many Research queries will I need?

This project uses three Research queries (one per persona). Free users get limited Research queries per day. Pro subscribers get extended access. Confirm you have at least three queries available before continuing.

ⓧ I don't see Research in the mode selector

Make sure you're signed in. Research mode requires an active Perplexity account.
Try refreshing the page. The mode selector sometimes takes a moment to load.
Confirm you're on perplexity.ai (not a cached or outdated bookmark).

Still can't find Research?

Research is available to both free and Pro users. If the option doesn't appear after signing in and refreshing, try clearing your browser cache or opening Perplexity in an incognito window.

Help me find Research mode in Perplexity

What does Research mode do?

Research mode uses search and coding capabilities to iteratively search, read documents, and reason about next steps. It refines its research plan as it learns more about the subject.

After evaluating source materials, it synthesizes everything into a clear, comprehensive report. Most reports complete in under 3 minutes, though complex topics can take 4 to 5 minutes.

Prepare your startup materials

Your startup idea is the consistent input you'll feed to each reviewer GPT. Using the same input across all three lets you compare their feedback directly and see how each persona evaluates the same information differently.

Gather whatever best describes your startup idea: a pitch deck, a one-pager, a product description, or a few paragraphs explaining the concept.
Save it in a place where you can easily copy and paste the full text, such as a text document or note-taking app.
Open your saved materials and confirm you can select and copy the full text of your startup description.

Keep your input consistent

The whole point of having three different reviewers is to compare their perspectives on the same idea. If you change or expand your input between reviewers, you won't be able to tell whether the differences come from the persona or from the input.

Your tools are verified and your materials are ready. Next up, you'll create your first reviewer GPT with basic instructions and see firsthand why generic AI feedback falls short.

Build and Test a Naive VC Reviewer

Your tools are verified and your startup materials are ready to go. Now it's time to build the first reviewer in your panel.

You'll create a VC Reviewer custom GPT using only your own instincts for what a venture capital investor would say. This first version establishes a baseline you'll measure against later.

In this step, get ready to:

Create a custom GPT with basic VC reviewer instructions.
Test it by pasting your startup materials and reading the feedback.
Save the output as your baseline for comparison.

Create the naive VC Reviewer GPT

The GPT builder in ChatGPT lets you create a standalone assistant with its own name, instructions, and conversation starters. You'll use the Configure view to set everything up directly.

Return to https://chatgpt.com/gpts in your browser.
Click Create to open the GPT builder.
Switch to the Configure view at the top of the builder.
In the Name field, enter Naive VC Reviewer.
In the Description field, enter Reviews startup ideas from a VC perspective..

Copy the following text into the Instructions field:

You are a venture capital investor. Review startup ideas and give feedback on whether they're investable. Be honest and direct.

What do these instructions do?

This tells the GPT to adopt a VC investor role and give direct feedback on whether a startup idea is worth investing in. That's it. No frameworks, no evaluation criteria, no specific metrics to check.

In the Conversation starters section, add Review my startup idea.
Add a second conversation starter: Is this startup investable?.
Click Create in the top right corner to save your GPT.

What are conversation starters?

Conversation starters are pre-written prompts that appear as clickable buttons when someone opens your GPT. They give users a quick way to begin a conversation without typing a prompt from scratch.

You should see your Naive VC Reviewer GPT open in a fresh chat window, ready for a conversation.

Don't see your GPT?

If the builder shows an error, confirm that your ChatGPT workspace supports GPT creation. New GPT creation requires a Business, Enterprise, or Edu workspace. Personal accounts (Free, Go, Plus, and Pro) cannot create new GPTs.

If you're on a personal account, consider using ChatGPT Projects as a fallback to store your persona instructions instead.

help me troubleshoot creating a custom GPT in ChatGPT

Test it with your startup idea

Your naive VC Reviewer is live. Time to put it through its paces with your actual startup materials.

Before you paste anything, take a moment to predict: what kind of feedback do you expect? Do you think it will reference specific evaluation frameworks, or ask pointed questions about your unit economics?

In the Naive VC Reviewer chat window, paste your full startup materials (the pitch deck text, one-pager, or product description you prepared in Step 1).
Press Enter to send.
Read the full response from top to bottom.

Now compare what you got against what you predicted. Ask yourself: does this feedback mention specific VC evaluation frameworks like TAM/SAM/SOM, unit economics, or competitive moats? Does it probe your burn rate or founder-market fit? Or does it give high-level advice that could apply to almost any startup?

The feedback almost certainly reads as generic. It stays surface-level, offers broad encouragement or caution, and avoids the pointed, framework-driven questions that a real investor would ask. That gap is the whole point of this step.

Why is the feedback so generic?

The instructions you gave the GPT said "be a VC investor" but never defined what that actually means in practice. Without specific evaluation criteria, mental models, or frameworks to apply, the GPT falls back on general knowledge about startups.

This is the core problem that research-backed synthetic personas solve. In the next step, you'll use Perplexity to uncover how real VCs evaluate deals, then encode that knowledge directly into the GPT's instructions.

Save the baseline output

You'll compare this response against the research-backed version you build in the next step. Having the exact text side by side makes the transformation impossible to miss.

Select all the text in the Naive VC Reviewer's response.
Copy it using Cmd+C (macOS) or Ctrl+C (Windows).
Paste it into a note, document, or text file where you can find it easily.

Label your baseline clearly

Add a heading like "Naive VC Reviewer Output" at the top of your saved text. When you have three different reviewer outputs later in the project, clear labels save you from scrolling and guessing which is which.

You've got your baseline locked in. In the next step, you'll use Perplexity Research mode to uncover how real venture capital investors actually evaluate startups, then rebuild this GPT from the ground up with that knowledge.

Research and Rebuild the VC Reviewer Persona

In Step 2, you built a VC Reviewer with a few lines of generic instructions and saw the result. The feedback was surface-level and could have applied to any startup.

Perplexity Research mode searches dozens of sources and produces a cited report on any topic. In this step, you'll use it to research how real VCs evaluate deals, then rebuild your GPT with those frameworks encoded into its instructions.

In this step, get ready to:

Research early-stage VC evaluation frameworks using Perplexity Research mode.
Rewrite your GPT's instructions with structured, research-backed prompts.
Retest with the same startup idea and compare the output to your naive baseline.

Research how VCs evaluate startups

Your naive GPT gave vague feedback because its instructions had no grounding in real VC evaluation methodology. Perplexity Research mode can surface the specific frameworks, metrics, and mental models that actual investors use.

Switch back to perplexity.ai in your browser.
Click the mode selector dropdown in the search box.
Select Research.
Paste the following query into the search box and submit it:

How do early-stage venture capital investors evaluate startup ideas? What specific frameworks, mental models, red flags, and evaluation criteria do VCs use when reviewing pitch decks and startup pitches? Include specific metrics they look for (TAM, CAC, LTV, burn rate, runway), common rejection reasons, the questions they ask founders, and how they assess founder-market fit and competitive moats.

What should I look for in the report?

Research mode takes 3 to 5 minutes to generate a comprehensive report. Once it finishes, read through it and identify these key elements:

The evaluation framework VCs use (team, market, product, traction, business model).
Specific metrics and thresholds (TAM/SAM/SOM, CAC, LTV, burn rate, runway).
Common red flags that lead to rejection.
The types of questions VCs ask founders during diligence.
How investors assess founder-market fit and competitive moats.

Once the report loads, you should see a detailed, multi-section analysis with cited sources covering VC evaluation criteria.

Report not generating?

Confirm you selected Research from the mode selector dropdown, not a different mode. Free users get limited Research queries per day. If you've hit your limit, wait until the next day or upgrade to Perplexity Pro for extended access.

Still stuck? Ask for help with Perplexity Research mode.

Rewrite the GPT instructions from research

Now you'll transform your Perplexity findings into structured GPT instructions. The goal is to encode the frameworks, metrics, and behavioral patterns you just researched into a prompt that makes your GPT reason like an actual early-stage investor.

Switch back to ChatGPT in your browser.
Return to https://chatgpt.com/gpts and find your Naive VC Reviewer GPT.
Select the edit (pencil) icon to open the GPT builder.
Switch to the Configuration view.
Replace the current text in the Instructions field with a new structured prompt organized into four sections:

Role and perspective defines who this persona is. Describe an experienced early-stage investor, their typical investment thesis, and their evaluation mindset.
Evaluation framework lists the specific criteria they assess. Pull directly from your Perplexity research: team, market size, product differentiation, traction signals, and business model viability. Include the specific metrics and thresholds your research identified.
Response format defines how the GPT structures its feedback. Specify sections like a summary assessment, framework-by-framework analysis, red flags, and follow-up questions.
Behavioral rules set what the GPT always probes, what it flags as dealbreakers, and its tone. Draw from the decision-making patterns and red flags in your research.

Why structure the instructions this way?

OpenAI's own guidance recommends using headings and lists so priorities and steps are visually distinct. Positive, concrete instructions ("Always probe unit economics") outperform long lists of prohibitions ("Don't give vague feedback").

Separating role, framework, format, and behavior gives the GPT clear guardrails without conflicting instructions. Each section answers a different question: who are you, what do you assess, how do you present findings, and what rules do you follow.

💡 Upload your research as a Knowledge file

You can export your Perplexity report and upload it to the Knowledge section of your GPT. This lets the GPT reference the full research directly, giving it deeper grounding beyond what you summarized in the instructions.

Update the Name field from Naive VC Reviewer to VC Reviewer.
Select Update to save your changes.

Don't see the Update button?

Make sure you're in the Configuration view, not the conversational builder. Changes save to a draft automatically while you edit, but you need to select Update to apply your latest draft changes.

Still having trouble? Ask for help updating your GPT.

Retest and compare against your baseline

The real test of a research-backed persona is whether it produces materially different output from the naive version. You saved the naive output in Step 2 for exactly this moment.

Before you test, make a prediction. Do you think the updated VC Reviewer will reference specific evaluation frameworks and ask pointed follow-up questions about your startup's metrics?

Open your updated VC Reviewer GPT and start a new conversation.
Paste the exact same startup materials you used in Step 2.
Read the response carefully before scrolling past it.

The response should now reference specific evaluation frameworks like market sizing, unit economics, and founder-market fit. It should ask pointed follow-up questions and give feedback that feels specific to your startup.

Pull up the naive output you saved in Step 2.
Read both responses side by side and note the specific differences.

What should the difference look like?

The naive version likely gave broad encouragement or surface-level concerns. The research-backed version should dig into specifics: TAM/SAM/SOM analysis, CAC-to-LTV ratios, burn rate assumptions, competitive moat assessment, and founder-market fit.

If both versions feel similar, your instructions may need more specificity. Go back to the Instructions field and add more concrete frameworks, metrics, and behavioral rules drawn from your Perplexity research.

You've just experienced the core methodology of this project: research a real-world archetype, extract its frameworks, and encode them as structured GPT instructions. Next up, you'll build a Startup Sparring Partner that stress-tests your thinking and challenges your assumptions.

Research and Build the Startup Sparring Partner

Your VC Reviewer now delivers research-grounded feedback targeting specific metrics and evaluation criteria. That is a powerful evaluation lens, but it is only one lens.

Founders also need someone to stress-test their thinking and challenge assumptions before investing time and money. A Sparring Partner plays a fundamentally different role: adversarial coach, not judge.

In this step, get ready to:

Research how experienced startup advisors challenge founders' assumptions using Perplexity Research mode.
Build a Startup Sparring Partner GPT with research-backed instructions that stress-test your thinking.
Test the Sparring Partner against your startup idea and contrast its output with the VC Reviewer's feedback.

Research the Sparring Partner persona

The VC Reviewer evaluates whether an idea is investable. A sparring partner evaluates whether the founder's thinking holds up under pressure. To build a persona that coaches that way, you need to understand the specific techniques real advisors use.

Switch back to Perplexity in your browser.
Select Research from the mode selector in the search box.
Paste the following query into the search box and submit it:

How do experienced startup advisors and mentors challenge founders' assumptions and stress-test business ideas? What specific techniques do they use, such as the Five Whys, assumption mapping, pre-mortem analysis, and devil's advocate questioning? How do advisors at accelerators push founders to think more critically? Include the tone, pacing, and conversational style of effective adversarial coaching versus destructive criticism.

Why a separate research query?

You already ran a Perplexity Research query for the VC persona, but that one focused on investment evaluation. This query targets a completely different domain: coaching methodology and adversarial questioning techniques.

Each persona gets its own research because each models a different real-world archetype. Reusing the same research would produce personas that all sound the same.

Wait for the report to generate. This typically takes 4 to 5 minutes.
Read through the full report and identify these key elements: questioning techniques and frameworks, assumption-challenging methods, the tone of constructive adversarial coaching, and how sparring differs from evaluation.

Save or export the report

Keep this report open or export it to a document. You will pull specific techniques directly from it when writing your GPT instructions in the next substep.

🙋‍♀️ Research mode not available?

Free Perplexity users get limited Research queries per day. If you have used your daily allowance, wait until the next day or upgrade to Pro for extended access.

Help me troubleshoot Perplexity Research mode access for this project.

Build the Sparring Partner GPT

With your research in hand, you can now create a custom GPT that embodies the adversarial coaching style you just studied. The goal is to translate those techniques into structured behavioral instructions.

Return to ChatGPT and visit https://chatgpt.com/gpts.
Select Create to open the GPT builder.
Switch to the Configuration view.
Set the Name to Startup Sparring Partner.
Add a Description such as Challenges startup assumptions and stress-tests founder thinking.

Now for the most important field. In the Instructions box, write structured instructions grounded in your Perplexity research. Use the same methodology you used for the VC Reviewer: translate real-world expertise into behavioral rules.

✔️ I'll give it a go!

Structure your instructions with these four sections, filling in each with findings from your Perplexity report:

Role and perspective - Define the persona as an experienced advisor whose job is to challenge thinking, not evaluate investability.
Questioning techniques - List the specific techniques from your research (Five Whys, assumption mapping, pre-mortem analysis, devil's advocate questioning, and others you found).
Tone and interaction style - Set the tone to direct but constructive, using the pacing and conversational style insights from your research.
Pushback structure - Define how the persona structures its challenges (identify the strongest assumption, ask what would have to be true, probe the failure scenario).

ⓧ Maybe a little help...

Use this template as a starting point. Replace the guidance notes in parentheses with the specific findings from your Perplexity research.

## Role and Perspective
You are an experienced startup advisor and sparring partner. Your job is to challenge the founder's thinking, not evaluate investability. You act as an adversarial coach who pushes founders to think more critically about their assumptions.

## Questioning Techniques
Use the following techniques to challenge the founder's thinking:
(List the specific methods from your Perplexity research here, e.g. Five Whys, assumption mapping, pre-mortem analysis, devil's advocate questioning.)

## Tone and Interaction Style
Be direct but constructive. Push back firmly without being destructive.
(Add pacing and conversational style details from your research here.)

## Pushback Structure
When reviewing a startup idea:
1. Identify the founder's strongest or most critical assumption.
2. Ask what would have to be true for that assumption to hold.
3. Probe the failure scenario if that assumption is wrong.
4. Surface hidden dependencies and second-order risks.
5. Move to the next assumption and repeat.

What makes this different from the VC Reviewer?

Notice the instructions never mention investment metrics, market sizing, or unit economics. The Sparring Partner focuses entirely on logic, assumptions, and blind spots. That distinction is what produces genuinely different output from the same startup idea.

Scroll down to Conversation starters.
Add Challenge my startup assumptions as the first starter.
Add Stress-test this business idea as the second starter.

Upload your research as Knowledge

You can optionally export your Perplexity report and upload it under Knowledge in the GPT configuration. This lets the GPT reference your research sources directly during conversations, making its challenges more grounded.

Select Create in the top right to save your new GPT.
Confirm that the GPT appears with the name Startup Sparring Partner and opens a fresh conversation window.

GPT not saving?

Make sure the Name and Instructions fields are filled in. The GPT builder requires at least a name and instructions to save.

Help me troubleshoot creating a custom GPT in ChatGPT.

Test and contrast with the VC Reviewer

The real test of a synthetic persona is whether it produces feedback that could only come from that specific archetype. Sending the same startup idea to both reviewers reveals whether the Sparring Partner genuinely challenges your thinking or just rephrases what the VC Reviewer already said.

Before you paste your startup materials, pause and predict: what do you expect the Sparring Partner to focus on that the VC Reviewer did not?

Open your new Startup Sparring Partner GPT conversation.
Paste in the same startup materials you used with the VC Reviewer.
Read the full response before doing anything else.

The Sparring Partner's response should challenge specific assumptions in your idea. Look for "what if" and "why not" questions. It should push back on weak logic rather than scoring investability.

Pull up your VC Reviewer output from the previous step.
Compare the two responses side by side. Ask yourself: Does the Sparring Partner focus on logic, assumptions, and blind spots? Does it avoid investment metrics like TAM or unit economics? Does it feel like a different kind of conversation entirely?

What should the contrast look like?

The VC Reviewer evaluates your idea against investment criteria. It scores and rates. The Sparring Partner interrogates your reasoning. It asks what would have to be true for your plan to work, then probes what happens if it does not.

If both responses feel similar, your Sparring Partner instructions may be too close to the VC Reviewer's. Go back and sharpen the distinction by removing any evaluation-oriented language and strengthening the coaching and questioning techniques.

🙋‍♀️ Feedback feels too similar to the VC Reviewer?

Check your Instructions for any investment-oriented language like "evaluate," "score," "assess investability," or references to TAM, CAC, or LTV. Remove those and replace them with coaching verbs like "challenge," "probe," "question," and "stress-test."

Help me make my Sparring Partner GPT produce different feedback from my VC Reviewer.

You now have two research-backed personas that give genuinely different feedback on the same idea. Next up, you will build the third and final reviewer: a Customer Interview Simulator that roleplays as a potential buyer and surfaces objections you have not considered yet.

Research and Build the Customer Interview Simulator

Your VC Reviewer evaluates investability. Your Sparring Partner stress-tests your assumptions. Two powerful perspectives, but neither answers the question that ultimately determines success: will real customers actually buy this?

The perspective that matters most for product-market fit comes from the customer. In this step, you'll research how customers actually behave in discovery interviews and build a Customer Interview Simulator GPT that lets you rehearse those conversations and surface objections you haven't anticipated.

In this step, get ready to:

Research customer discovery interview dynamics using Perplexity Research mode.
Build a Customer Interview Simulator GPT with research-backed persona instructions.
Test a roleplay customer interview and compare the feedback with your other two reviewers.

Research customer interview dynamics

The VC Reviewer and Sparring Partner both give structured, analytical feedback. A real customer doesn't think that way.

Real customers hedge, give polite non-answers, and express reservations indirectly. Your research needs to capture those patterns so the simulator feels like a genuine conversation, not a chatbot spitting out bullet points.

Switch back to Perplexity in your browser.
Select Research from the mode selector in the search box.
Enter a query like the one below to research customer discovery dynamics:

How do target customers respond during startup customer discovery interviews? What common objections, hesitations, and buying barriers do early-stage customers raise? Include insights from The Mom Test methodology by Rob Fitzpatrick, jobs-to-be-done frameworks, and common patterns in customer interview transcripts. What makes customer responses realistic versus polite but useless? How do customers express unmet needs, price sensitivity, and switching costs?

Wait for the report to generate. This typically takes 4 to 5 minutes.

Once the report is ready, read through it carefully. You're looking for patterns that will make your simulator feel like a real person, not an AI assistant.

What to extract from this research

As you read the report, identify these specific patterns:

Common customer objection patterns like price sensitivity, switching costs, and status quo bias.
How real customers express genuine interest versus politeness bias. The Mom Test methodology is especially useful here.
Buying hesitation triggers that surface in real discovery interviews.
Realistic conversational patterns from jobs-to-be-done frameworks and real interview transcripts.
What separates a useful simulated interview from a compliant chatbot conversation.

🙋‍♀️ Research mode not available?

Free users get limited Research queries per day. If you've run out, you can wait until the next day to continue or upgrade to Perplexity Pro for extended access.

Need more help?

Help me troubleshoot Perplexity Research mode access.

Build the Customer Interview Simulator GPT

This GPT is fundamentally different from the other two. The VC Reviewer and Sparring Partner both analyze your idea. The Customer Interview Simulator roleplays as a real person.

Your instructions need to keep it in character at all times. If it slips into bullet-point analysis or structured feedback, it stops being a customer simulator and becomes another analyst.

Return to ChatGPT in your browser.
Go to https://chatgpt.com/gpts and select Create.
Switch to the Configuration view.
Set the Name to Customer Interview Simulator.
In the Instructions field, write a structured prompt grounded in your Perplexity research. Use the structure below as your guide.

How to structure the instructions

Use headings and sections to organize your instructions, just like you did for your other GPTs. Include these four areas, drawing directly from your research findings:

Customer persona: Define who this person is. Include their role, their current solution to the problem your startup solves, and their pain points and priorities.
Response style: Set the tone to realistic conversational language. The customer should respond in natural sentences, not bullet points or structured analysis.
Objections and hesitations: Draw from your research to include specific barriers like price sensitivity, switching costs, and status quo bias.
Interaction rules: The customer answers like a real person. They don't volunteer information unless asked. They express genuine reservations. They never break character or switch to an analytical mode.

Add these two conversation starters in the Conversation starters field:
Optionally, export your Perplexity report and upload it as a Knowledge file so the GPT can reference your research directly.
Select Create to save your GPT.

Your Customer Interview Simulator GPT is now live. You should see it in your GPTs list alongside the VC Reviewer and Startup Sparring Partner.

GPT not saving?

Make sure the Name and Instructions fields are filled in. A GPT needs at minimum a name and instructions to be created.

If the Create button is greyed out, check that you haven't accidentally switched to the conversational builder tab. Switch back to the Configuration view.

Help me troubleshoot creating my Customer Interview Simulator GPT.

Test a customer discovery interview

Time to see if your Customer Interview Simulator feels like a real conversation. Before you start, think about what you expect. Will this GPT give you structured feedback like the VC Reviewer and Sparring Partner did, or will the response feel fundamentally different?

Open your new Customer Interview Simulator GPT.
Paste your startup materials as if you're pitching to a real prospect.
Read the initial response carefully.

The response is different from anything your other two reviewers produced. The Customer Interview Simulator stays in character, responding as a real person would. It asks clarifying questions, expresses realistic hesitations, and raises objections grounded in real customer behavior patterns.

Continue the conversation with 2-3 follow-up questions, just like you would in a real customer interview.
Ask about the customer's current workflow, what frustrates them, and what would make them switch to your product.

How does this feedback differ from your other reviewers?

Compare across all three lenses. The VC Reviewer scored your idea against investment criteria. The Sparring Partner challenged your assumptions and poked holes in your logic.

The Customer Interview Simulator surfaced buying objections and unmet needs from the user's perspective. Each GPT gave you feedback the others couldn't, because each was built from different real-world research.

💡 What if the simulator breaks character?

If the GPT starts giving you bullet-point analysis or structured feedback instead of staying in character, your instructions need stronger behavioral rules. Go back to the Configuration view and add more explicit rules about never switching to an analytical mode.

Your customer simulator now gives you a research-backed rehearsal partner in ChatGPT. Next, you'll recreate it in Claude and compare how two models handle the same persona, evidence, and interview prompt.

Recreate the Customer Interview Simulator in Claude

Your Customer Interview Simulator now roleplays realistic buyer conversations as a custom GPT. Next, you'll rebuild the same simulator with Claude Projects so you can compare how two models handle the same persona and evidence.

A Claude Project keeps persona instructions and research available across its chats. This gives you a fair comparison because both simulators use the same behavioral rules and customer evidence.

In this step, get ready to:

Create a Claude Project for your Customer Interview Simulator.
Add your persona instructions and customer interview research.
Run the same interview in Claude and your custom GPT.

Create the Claude Project

Projects are available to every Claude user. Free accounts can create up to five projects, so you can complete this comparison without changing plans.

Open Claude Projects in your browser.
Select + New Project in the upper-right corner.

You'll see fields for the project's name and description.

Enter Customer Interview Simulator as the project name.
Enter Practice realistic customer discovery interviews using research-backed behavior. as the description.

If your Team or Enterprise account shows visibility options, keep the project private unless you want colleagues to use it.

Complete the project setup.

You should see the Customer Interview Simulator project page with its knowledge area on the right.

Where Claude gets its role

Claude uses the project name and description for organization. Put the simulator's behavior in project instructions because Claude does not use the name or description as chat context.

Files in project knowledge supply evidence across the project's chats. Instructions shape the role, while knowledge grounds the role in your research.

Add the persona and research

Your custom GPT already contains the persona you researched. Reusing those instructions keeps the model as the only major difference in this comparison.

Open your Customer Interview Simulator in ChatGPT.
Open its Configuration view.

You'll see the instructions that control the simulator's persona, response style, objections, and interaction rules.

Copy the complete contents of the Instructions field.
Return to your Customer Interview Simulator project in Claude.
Select Set project instructions.

You'll see the editor that controls how Claude behaves in every chat inside this project.

Paste the copied persona instructions into the editor.
Select Save instructions.

The saved instructions now give every new project chat the same customer role.

Open your customer interview report in Perplexity if you did not export it earlier.
Export the report as a PDF or document.

You should see the report in your browser's downloads.

Return to the project knowledge area in Claude.
Select the + button.
Upload the exported customer interview report.

That's the second simulator configured. Claude now has both the behavioral rules and the research evidence behind them.

Compare the two simulators

A fair comparison uses the same startup context and interview question in both tools. You can then focus on differences in tone, realism, and resistance.

Start a new chat inside your Customer Interview Simulator project.

Before you send the interview prompt, do you expect Claude to sound more cooperative, more skeptical, or about the same as your custom GPT?

Start the roleplay by sending this prompt:

Roleplay as the target customer for this startup: [[STARTUP_IDEA="describe your startup idea in one sentence"]]. I am conducting a customer discovery interview. Tell me how you handle this problem today, what frustrates you about the current approach, and what would make you hesitate to switch. Stay in character and answer in natural conversation.

What should a strong response show?

The customer speaks in first person instead of analyzing the startup from outside the role.
The answer sounds conversational instead of presenting a polished framework.
The customer names concrete habits, frustrations, or switching barriers.
The response leaves useful gaps for your next interview question.

Check Claude's response for the four signs above.
Return to your Customer Interview Simulator GPT in ChatGPT.

You'll see a fresh conversation ready for the comparison prompt.

Start a new conversation with the GPT.
Send the same completed interview prompt.

You should now have two responses to the same interview. Compare which simulator stays in character, reveals believable friction, and gives you stronger follow-up questions.

Claude breaking character?

Check that the project instructions explicitly require natural sentences and first-person answers.
Confirm that the Perplexity report appears in the project knowledge area.
Start a new project chat so earlier coaching prompts do not influence the comparison.

Help me keep my Claude customer simulator in character.

You now have the same research-backed customer simulator on two platforms. That comparison shows which model gives you the most useful rehearsal partner for your next real interview.

Secret mission

Build a Review Panel Synthesis GPT

You have three reviewers giving three distinct perspectives on your startup idea. But no founder wants to reconcile three separate reports manually. Can you build a fourth GPT that synthesizes all three outputs into a single, prioritized action plan with confidence levels?

Clean Up Your Resources

Decide whether to keep your resources running, pause them to come back later, or delete them entirely. This project uses your existing ChatGPT and Perplexity subscriptions, so there are no ongoing costs from the resources you created.

Resources you used:

VC Reviewer custom GPT (hosted in your ChatGPT workspace).
Startup Sparring Partner custom GPT (hosted in your ChatGPT workspace).
Customer Interview Simulator custom GPT (hosted in your ChatGPT workspace).
Review Panel Synthesis custom GPT (hosted in your ChatGPT workspace, Secret Mission).
Four Perplexity Research threads (VC evaluation, sparring techniques, customer interviews, synthesis frameworks).
Saved naive VC Reviewer output (your baseline comparison text).

Keep everything running

No action needed. Choose this if you plan to keep using your Review Panel to evaluate new startup ideas or refine existing ones.

All four custom GPTs remain in your ChatGPT workspace, ready to use whenever you start a new conversation with them.
Your Perplexity Research threads stay in your thread history for reference.
Your saved naive VC output stays wherever you stored it for future comparison.
None of these resources incur ongoing costs. They sit idle until you use them again.

Pause - I'll come back to this later

Your GPTs and Perplexity threads have no running processes, so there is nothing to shut down. Everything stays in your accounts at no cost until you return.

Your four custom GPTs remain saved in your ChatGPT workspace. Each conversation with a GPT starts fresh, so there is no state to pause.
Your Perplexity Research threads stay accessible in your thread history whenever you need to revisit the research.
Keep your saved naive VC output. It is useful for demonstrating the before-and-after impact of research-backed prompts.

Delete - I don't want to use this again

Remove all project resources from your accounts. Follow each item below to delete everything you created.

Delete each custom GPT. Visit https://chatgpt.com/gpts and locate your GPTs. For each one (VC Reviewer, Startup Sparring Partner, Customer Interview Simulator, and Review Panel Synthesis), open the GPT and click the edit (pencil) icon. In the editor, look for the delete option in the menu to remove it permanently.
Delete your Perplexity Research threads. In Perplexity, locate each Research thread in your sidebar. Hover over the thread title and select the option to delete it. Repeat for all four research threads (VC evaluation, sparring techniques, customer interviews, synthesis frameworks).
Delete the saved naive VC Reviewer output from wherever you stored it (a notes app, a document, or a text file on your Desktop).

Nice Work!

Nice work! You just built a complete Startup Review Panel powered by research-backed synthetic personas.

What you learned:

Built three research-backed synthetic reviewer personas as custom GPTs, each providing a distinct evaluation lens on your startup idea: a VC assessing investability, an advisor stress-testing assumptions, and a customer surfacing buying objections.
Owned a repeatable persona-building methodology. Research the real-world archetype with Perplexity, extract frameworks and mental models, then encode them as structured behavioral instructions in a custom GPT.
Experienced the difference between naive and research-backed AI instructions firsthand by comparing generic feedback against specific, framework-driven evaluation on the same startup idea.
Secret Mission: Built a Review Panel Synthesis GPT that consolidates multi-persona feedback into a prioritized startup action plan with confidence levels.

Ready to quiz yourself?