Competitive Research Using AI (Without The Hallucinations) For PMs
- Harshal Patil
- Jun 6
- 7 min read
Step-By-Step Guide And Real Example To Inform Noise-Free, Confident Feature Decisions
When you think about a new feature for your product, apart from user research, research competing products. This helps understand how other product teams solve the same user problem. It’s not about copying features (which may not succeed) — it’s about learning how others address similar needs and what users expect.
Here is a repeatable process that helps you define features better and faster.
The Ideal Workflow
In an ideal world, you would ask an AI tool (e.g., ChatGPT deep research) to read a JIRA ticket, a customer escalation, or a sales conversation. AI would respond immediately with a full competitive analysis, complete with product screenshots, pricing tables, GTM positioning, user quotes, and citations. But that’s not how things work yet.
A Realistic Workflow, Today
Here’s the step-by-step process you can follow right away:
1: Define the landscape
2: Create the research prompt
3: Run research queries
4: Verify the sources
5: Extract evidence from each source
6: Summarize and synthesize
Step 1: Define The Landscape
1.1: Define the user need or JTBD
What is the core problem we’re solving?
1.2: Collect jargon
What do users or the industry call this feature or need? You need to know the jargon and alternative phrasing of the feature or need.
1.3: Identify direct competitors
Which products share a similar architecture or value proposition? These are direct competitors.
1.4: Identify indirect competitors
Who solves the same user problem using a different approach? These are indirect competitors as they meet the same Job to be done (JTBD)
Step 2: Create A Research Prompt
Feed your answers from Step 1 to AI. Ask it to craft a single, detailed prompt you can reuse for competitive analysis.
Step 3: Run Research Queries
Use two tools for your research. Assign fewer complex queries to one, and give smaller or parallel queries to the other.
1 – Start with ChatGPT Deep Research. Feed it the full research prompt to generate a comprehensive overview.
2 – Take the same prompt (or simplify it, e.g., “How do I do X using Y tool?”) and run it through Perplexity to get a diverse set of sources and perspectives.
Step 4: Verify Sources
Review each output. Open the “sources” and confirm that the cited references discuss the actual competitor product and feature you're researching.
Exclude Mismatched Sources
If a citation points to unrelated tools or products, eliminate that competitor from your list for that specific feature. AI can hallucinate—claiming that competitor X supports feature Y while citing sources that only discuss competitors A, B, and C. Trust the cited references, not the AI summary.
Forums and blogs
Not all information comes from official websites. Some helpful insights come from user-generated content (UGC), including forums or technical blogs.
Step 5: Extract Evidence From Each Source
Open each reference link individually and read through it carefully. Extract the most relevant information and save it in a Notion or Google Doc.
Include the URL so you can revisit the source later or cite it if needed.
As a shortcut, you could paste each webpage into a new chat with an AI assistant to parse it for relevant insights. However, reading builds understanding. Be mindful that this shortcut may short-circuit the deeper learning that comes from reading and interpreting the material yourself.
Step 6: Summarize and Synthesize
Once you’ve gathered notes on competing products—whether it's 5 or 30—use an AI tool like Notion AI to summarize them into a few clear bullet points per area. Focus on patterns and distinctions to synthesize insights effectively. Some synthesis examples:
How many competitors support feature X? How many don’t?
Supported use cases: Text, image, video, etc.
Pricing strategies per product.
As you review and synthesize, apply simple heuristics to focus your analysis. For example, use Kano model to ask if this feature is a must-have, a differentiator, or a performance need?
Once synthesized, these insights feed directly into writing product specs—helping you define requirements, differentiate, and avoid guesswork.
Loop, Not a Line
Treat Steps 1–6 as a cycle, not a straight line: each round of synthesis should prompt you to revisit your JTBD, competitor list, or research prompts. As you uncover new patterns or edge cases, loop back to redefine the problem or expand your competitor set before locking in feature requirements.
Example: An AI Meeting Notetaker Feature
1.1: What is the core problem we’re solving? (example)
We want to add support for not just virtual Google Meets and Zoom meetings but also for Hybrid, In-Person Meetings, Gather, Slack, or phone calls.
The core JTBD is that users don’t just have virtual meetings—they make decisions and exchange critical information across in-person meetings, Slack huddles, or phone calls. Their need remains the same: capture what happened, remember what matters, and ensure follow-up. We’re expanding support to ensure that no meeting—regardless of format—is left undocumented or un-actioned. Participants want confidence that their notes and action items are accessible and shareable across any meeting tool or setting, reducing stress and ensuring everyone—attendees and absentees alike—can stay informed and contribute.
1.2: What do users or the industry call this feature or need? (example)
Prompt:
I have an AI meeting notetaker product. I want to add support for not just virtual Google Meets and Zoom meetings but also for Hybrid, In-Person Meetings, Gather, Slack, or phone calls. What is this capability called in the industry?
Output excerpt:
Integration with multiple platforms, Cross-Platform Meeting Capture, Multi-Modal Meeting Transcription, in-person meeting transcription, in-person note-taking
1.3: Direct competitors (example)
Prompt:
Which AI meeting notetakers can support Hybrid, In-Person Meetings, Gather, Slack, or phone calls?
Alternative prompt, add this to the above:
include those with Integration with multiple platforms, Cross-Platform Meeting Capture, Multi-Modal Meeting Transcription, in-person meeting transcription, and in-person note-taking
Output excerpt:
ScreenApp
Fireflies AI
Fellow
Notta
Krisp
Otter AI
Claap
Jamie AI
Granola
1.4: Indirect competitors (example)
Prompt after sharing JTBD:
for this JTBD, what are other ways to solve it apart from AI notetakers
Output:
Google Docs, Notion, and Confluence meeting note templates
Full-time secretary
Checklists or task trackers like Trello or Asana.
Whiteboard photos
Rotating designated scribe
2: Create The Research Prompt (example)
Prompt:
Based on this information, can you draft a deep research prompt? The goal of the research is to understand how competitors do it. So I can figure out how we should do it.
Output excerpt:
Understand how competing tools implement support for non-virtual meetings (e.g., hybrid, in-person, phone, Slack Huddles) so we can define how our product should build similar or better capabilities.
For each of the following competitors, investigate: Do they support in-person or hybrid meeting capture? How do they do it? How do users start capturing an in-person or phone call meeting? What input method is used (e.g., mobile voice record, audio file upload, Slack bot)?
How are action items or notes generated and shared afterward? What permissions or privacy options are given?
Do they integrate across platforms (Slack, Zoom, Meet, MS Teams)? How do they merge notes across formats (e.g., hybrid: in-room + Zoom)? Do they support syncing with project management tools? Is this capability gated behind a specific plan?
List of competitors given.
3: Run Research Queries (example)
1 - Above big prompt with ChatGPT Deep Research
2 - Small prompt examples with Perplexity:
How do I record a slack huddle using fireflies AI?
How do I record an in-person meeting using Granola?
How does Fellow identify the speaker when in an unstructured or in-person meeting?
4: Verify the sources (example)
Check the sources the AI used to generate its answer. Are the sources related to the product you asked? Or does the answer seem coherent, but the sources are of other products?
For example, most of the sources used by Perplexity for the below prompt are not related to NotyAI.
How is notyai recording google meets
5: Extract evidence (example)
Read through the linked webpages to make your mind. Make notes as you see fit. I copy over these into one doc:
The webpage URL
Text excerpts
Code excerpts
Images
6: Synthesize (example)
Example prompts:
Make a comparison table using the notes of each product with one row per product and one column per channel supported, including in-person meetings, Zoom, Google Meet, and so on.
Write a JTBD for a PRD to build a feature to meet the user needs as described in this competitive research. Anchor the JTBD to what’s feasible. Keep technical implementation details separate. Where unclear, ask clarifying questions to direct my further research.
For each competitor in my list, draft a one-page GTM cheat sheet that covers: target persona(s), core value proposition, primary pricing tiers, messaging pillars, and any notable channel or partnership strategies.
Synthesis helps you learn and internalize the research you conducted. So, I suggest comparing the synthesis with your notes.
GenAI will help you overcome the blank slate, but not complete the job for you.
Process Challenges
Even though this process works well, there are real challenges.
Validate AI Output: Don’t trust AI blindly. Use it to speed up research, but always check the original sources. AI connects links into a story better than a Google search, especially with tools like Perplexity—but watch out for hallucinations. If it claims “Competitor Y solves Need X,” verify that the sources actually support that.
Invest the Time: Accept that there’s no true shortcut. AI accelerates the process, but competitive research still takes time. Plan accordingly. You get what you put into it.
Prioritize Text Over Images: Let AI help with summarizing and comparing text, but manually review diagrams and flowcharts. AI struggles to accurately synthesize visual information.
Work Around Hidden Pricing: Understand that enterprise pricing is rarely public. Don’t expect AI or search engines to produce exact numbers. Instead, piece together benchmarks from forums, user anecdotes, and scattered clues that AI can gather.
Stay Grounded in User Needs: Avoid the trap of building just because a competitor does. Anchor your roadmap in user interviews and support tickets. When direct user data is limited, use competitor research to uncover patterns and broaden your view—without losing sight of what your users actually ask for. Extrapolate from a few users of your product to learnings from the countless users of all your competitors put together.
If you have any questions or want help analyzing your competitors, get in touch.