← Back to blog
·13 min read
Written by:
JR
Jordan Reyes
Verified by:
CL
Casey Lin

How PainPointMap Scores Pain Points (Methodology)

How PainPointMap collects Reddit posts, extracts pain points, scores severity and frequency, grounds citations and maps gaps, plus what the scores are not.

Share:

Key Takeaways

  • Each scan pulls recent posts from the subreddits you pick, ranks pain-signal posts first, and sends at most 50 posts to the model.
  • Severity is a 1 to 100 model rating of frustration, and frequency is the model count of posts that mention the problem.
  • Every cited source ID is checked against the posts actually collected, and any ID that does not match is deleted before saving.
  • Competitor mapping and gap scoring run only on Professional, and the competitor list comes from model knowledge, not a live web check.
  • Scores rank problems inside one scan. They are not demand measurements, and they cover only the sampled Reddit window.

Quick answer: PainPointMap pulls a window of recent posts from the subreddits you choose, ranks posts that contain pain-signal language, and has an AI model extract pain points with a 1 to 100 severity rating, a frequency count and cited source threads. Every citation is checked against the posts actually collected, and the scores are relative judgements, not demand measurements.

This post describes how a scan works, step by step, including the places where the tool is capped or approximate. It tries to say plainly what the product does and what it cannot do.

If you want a version you can run by hand, the sibling post on a pain point scoring model you can run yourself covers that. For the broader discipline, start with what is pain point research.

The Pipeline at a Glance

A scan runs in a fixed order. Each stage has a job and a limit.

StageWhat happensLimit worth knowing
1. CollectRecent posts are pulled for each subreddit in your chosen time rangeTime range is 7 days, 30 days, 90 days or 1 year
2. RankPosts with pain-signal phrases go first, then high-scoring posts fill the restUp to 100 posts kept per subreddit
3. Add commentsComments are fetched for the top-ranked posts onlyA handful of posts per subreddit, fewer when you pick more subreddits
4. ExtractAn AI model groups complaints into pain points and rates themAt most 50 posts go to the model
5. GroundCited source IDs are checked against collected postsUnmatched IDs are removed
6. EnrichSolution ideas on every plan, competitor mapping on ProfessionalUp to 3 ideas on Starter, up to 10 on Professional
7. VerdictOne label and a short summary for the whole scanBased on the top ten pain points only

The rest of this post walks through each stage.

Where the Data Comes From

PainPointMap does not use Reddit's own interface. It reads from a public archive of Reddit posts and comments that updates with a short delay, and it paces its requests so the archive is not hammered. For context on why tools choose archives and what trade-offs follow, see our comparison of best Reddit scraping tools.

For each subreddit, a scan walks backward from now through the time range you selected, 100 posts per request, until it reaches the edge of the range or runs out of pages. The number of pages is capped, and the cap shrinks as you add subreddits so a five-subreddit Professional scan does not spend its whole time budget collecting. In practice that means a busy subreddit over a long range can contain far more threads than the scan ever sees. The sample is recent posts, not the whole history.

If the data source returns nothing, the scan fails and says so. It does not hand the model an empty list and let it improvise. A failed scan also gives your daily scan back. Failing loudly is a deliberate design choice, because a tool that invents findings from no data would be worse than no tool.

How Posts Are Selected for Analysis

Once the window of posts is collected, two things happen.

First, posts are split into two groups. The first group contains any post whose title or body includes a pain-signal phrase. The default list covers phrases such as "frustrated", "struggling", "wish there was", "alternative to", "problem with", "tired of" and "any recommendations". You can supply your own keywords, and those replace the defaults. The second group is everything else.

Second, each group is sorted by upvote score, and the two are joined with the pain-signal group first. The top 100 per subreddit are kept. So a quiet window still returns the subreddit's most discussed threads, but anything that sounds like a complaint is placed ahead of them.

Two honest consequences follow. Keyword matching is literal text matching, so a person who describes a problem without any of those phrases can rank lower. And ranking by upvotes favours posts that already got attention, which is not the same as posts that describe the sharpest problem. Our guide to Reddit research methods covers manual searches that catch what a phrase list misses.

Then comments are added. Comments are fetched only for the top-ranked posts, because each fetch is a separate request. When you scan one subreddit, more posts get comments than when you scan five. For each of those posts, the highest-scoring comments are kept.

Extraction: What the Model Actually Sees

Here is the limit that matters most. At most 50 posts per scan are sent to the model. Each post arrives as its ID, subreddit, title, score and comment count, the first part of the post body, and up to three comments, each trimmed to a short excerpt. The cap applies to the combined list, so with several subreddits the posts are not guaranteed to be split evenly between them.

The model is asked to group similar complaints into single pain points and, for each one, return a title, a short description, a severity rating, a count of posts that mention it, a category, keywords, short sample quotes and the IDs of the posts it came from. The posts are wrapped as data to analyze, and the model is told to ignore any instructions that appear inside them, which blunts attempts to inject text into a scan through a Reddit post.

Everything the model returns is treated as untrusted input. Titles and descriptions are trimmed to fixed lengths, scores are forced into valid ranges, malformed entries are dropped, and the number of quotes and keywords per finding is capped. That cleanup is about stability, not truth. It stops a broken response from corrupting a scan. It does not verify that a finding is correct.

Severity and Frequency Scores

Each pain point gets two main numbers.

Severity is a rating from 1 to 100 that the model assigns based on how frustrated the people in the sample sound. A refund dispute that blocks someone's work reads higher than a mild wish for a nicer interface. If the model returns a missing or invalid value, the stored default is 50. Severity is judgement, not arithmetic. Two runs over the same posts can produce slightly different numbers.

Frequency is the model's count of how many posts in the sample refer to that pain point. It is not a count of Reddit users and not a count across all of Reddit. A frequency of 12 means the model attributed roughly a dozen of the posts it saw to that problem. Because at most 50 posts are read, frequency has a natural ceiling per scan, and a count is only comparable between pain points inside the same scan.

Pain points are ranked by the model on severity and frequency combined. There is no published formula that multiplies the two. If you want a transparent, adjustable formula, that is what the manual scoring model is for, and our older guide on how to prioritize pain points explains the thinking behind weighing the two.

Citation Grounding: Why Source Links Can Be Trusted

Asking a model to cite sources is easy. Trusting the citations is not. Models invent plausible-looking identifiers when they are unsure, so the pipeline never takes a citation at face value.

Two things make citations usable. First, the real post ID is placed inside the prompt next to each post, so the model can copy an ID rather than make one up. Second, after the model answers, every cited ID is compared with the set of posts that were actually collected for that scan. Any ID that does not match a collected post is stripped out before anything is saved. What remains is mapped to the stored thread, which is why you can click from a finding to the original discussion.

This check proves that a cited thread exists in the sample. It does not prove the thread says what the finding claims. The model can still attach a real thread to a loosely related problem. That is why the right workflow is to open two or three cited threads per finding and read them. If you want a method for that reading step, see how to analyze Reddit data.

Competitor Mapping and Gap Scoring (Professional)

On the Professional plan, each pain point goes through a second pass. The model is asked to name two to five existing products that attempt to solve the problem, and for each one to give a short description, strengths, weaknesses, pricing if known, a gap note on what is still missing, and a relevance score from 1 to 100.

Be clear about the source. This step is generated from the model's own knowledge. It is not a live crawl of each vendor's site, so pricing and features can be out of date or wrong. Treat the list as a prompt for your own competitor research and verify anything you plan to rely on. Our guides on competitor gap analysis and the market gap analysis framework show how to do that check properly.

The competitor summaries then feed the solution step, where each idea receives a market gap score from 1 to 100 for how underserved the need appears. On Starter there is no competitor mapping, so solution ideas are generated without that competitive context. Gap scores are therefore a Professional feature in a real sense, and even there they are one model's estimate.

Solution Ideas

For each pain point, the model generates distinct solution ideas and is asked to vary the angle: a software feature, a standalone app, a browser extension, a service or a community, with different audiences and monetization models. Each idea includes a name, a description, a target audience, a monetization suggestion, a build difficulty and the gap score above. Starter returns up to 3 ideas per pain point and Professional up to 10.

These are brainstorming prompts, ranked best-first by the model's own gap score. They are not validated products. The useful move is to pick the two that surprise you and ask whether anyone in the cited threads is already paying for a workaround. If you want a route from idea to experiment, the seven-day Reddit validation plan is a good next read.

Start your 3-day free trial if you would like to see a full scan output. A card is required, you get 1 scan a day during the trial, and it then bills $19 (Starter) or $49 (Professional) a month unless you cancel.

The Verdict Label

At the end, the scan gets one label and a two to three sentence summary. The label is one of five: Strong Signal, Validated Demand, Mixed Signal, Weak Signal or Saturated Market.

The model writes it after reading a short list of the top ten pain points, each with only its title, severity and frequency. It does not re-read the posts, and it sees no revenue, search volume or payment data. If the scan finds no pain points, the label is Weak Signal. If the verdict step cannot complete in time, the label is derived from average severity instead: 70 or above is Strong Signal, 45 or above is Mixed Signal, and anything lower is Weak Signal. The summary says when that fallback was used, and it also notes if solution ideas or competitor maps were skipped for some pain points because the scan ran short on time.

The word "Validated" in "Validated Demand" is the model's category name. It is not an external validation. A label is a quick reading of the scan, and the right response to any label is more evidence, not a decision.

What the Scores Are Not

This is the section most methodology pages skip.

  • Not demand measurements. A high severity means people sounded frustrated in the sample. It says nothing about how many would pay.
  • Not a representative sample. Reddit users skew toward certain demographics and topics, and the sample is a recent window of only the subreddits you chose. Our post on Reddit data limits goes deeper, and Reddit vs surveys compares it with asking people directly.
  • Not comparable across scans. A 70 in one subreddit and window is not the same unit as a 70 in another, because the model rates relative to the posts in front of it.
  • Not stable to the decimal. The model's judgement can shift between runs.
  • Not a replacement for reading. Citations let you check the work. They do not do it for you.
  • Not outside Reddit. Nothing here sees Google searches, app reviews, support tickets or sales calls.

Used well, the output is a ranked, cited shortlist of problems worth a closer look. Used badly, it is a number that feels like proof. Our list of Reddit research mistakes describes the usual ways that goes wrong, and manual vs automated Reddit research explains when doing it by hand is the better call.

How to Use the Output Well

  1. Scan a narrow set of subreddits with a time range that matches how fast the topic moves.
  2. Read the cited threads behind your top three findings before trusting the scores.
  3. Cross-check each finding against a second source, such as customer interviews or a startup idea validation scorecard.
  4. Verify any competitor or pricing detail on the vendor's own site.
  5. Re-scan later with a different time range and see which problems persist.

If you are comparing tools rather than studying one, how to choose a Reddit research tool lists the questions to ask any vendor, including this one. For a fair side-by-side with a similar product, see PainOnSocial vs PainPointMap. Hand-rolled prompts in a chat window are another route, with their own limits, covered in using ChatGPT to analyze Reddit threads.

Final Thoughts

PainPointMap's scoring is a sampled, AI-judged ranking with checked citations. It is strongest as a way to find and rank candidate problems quickly, and weakest as a measure of how big or how profitable any of them is. The honest way to use it is to treat each score as a reason to read a thread, not a reason to build.

When you want to see how this works on your own subreddits, start your 3-day free trial. A card is required, the trial allows 1 scan a day, and after it ends the plan is $19 (Starter) or $49 (Professional) a month unless you cancel. More guides sit in the Reddit research guides hub, and the niche ideas by platform page can help you choose where to point your first scan.

Frequently Asked Questions

How does PainPointMap score pain points?

An AI model reads a sample of recent Reddit posts and the top comments on the most relevant ones, groups similar complaints, and gives each pain point a severity from 1 to 100 and a frequency count of posts that mention it. The scores come from model judgement over that sample, so treat them as a ranking inside one scan, not a market measurement.

Are the pain points in a scan real or made up?

The pain points are extracted from posts that were collected for that scan, and each cited source ID is checked against those posts. Any ID that does not match a collected post is removed. If no posts can be retrieved, the scan fails instead of producing findings. The model can still misread a thread, so open the cited links before you rely on a finding.

Does a Strong Signal verdict mean people will pay for a solution?

No. The verdict label is the model reading a short list of pain point titles with their severity and frequency, then judging whether the market looks worth pursuing. It does not see payment data, search volume or competitor revenue. Treat it as a first opinion to test with interviews and small experiments, not proof of demand.

How many posts does one scan analyze?

A scan walks back through the chosen time window for each subreddit, ranks the posts it finds, and sends at most 50 posts to the model, with up to three comments on each of the posts that had comments fetched. That is a sample, not every thread. Narrower subreddit choices and clear keywords give the sample a better chance of being relevant.

What does PainPointMap not measure?

It does not measure search demand, willingness to pay, market size, or anything outside Reddit. It also covers only the time range you pick, and Reddit users are not a random sample of customers. The scores are relative judgements inside one scan. Use them to decide what to investigate next, then confirm with other evidence.

Find what your customers actually complain about.

Scan the communities where your customers talk and see real pain points ranked by how often they come up — no surveys, no guesswork.

Scan My Market: 3-Day Trial
JR
Jordan Reyes
Research Writer, PainPointMap

Writes about Reddit market research, idea validation, and finding product opportunities worth building. Covers the niche and industry research guides on the blog.