# What a Buying-Intent Post Actually Looks Like: 1,416 Labelled Public Posts

> Buying intent signals from 1,416 labelled public posts: only 1 in 17 keyword matches was a lead. The phrases that predict buyers, false positives, and a rubric.

Canonical: https://useneedle.net/blog/buying-intent-posts-1416-labelled-examples
Published: 2026-11-21

Only about **1 in 17** public posts that matched a product's keywords was worth replying to, and only **1 in 67** was someone ready to buy. In the uniformly random part of the sample it was worse: 1 in 33. Most of what keyword matching returns is off-topic (53%) or people selling their own thing (30%).

I wanted to know what separates the 1 from the 16, using real posts instead of the generic "watch for frustration and competitor mentions" advice. So this post is built from a labelled dataset: 1,416 public posts, each matched to the keyword list of one of 202 products, each given a tier. Below are the counts, the phrases that actually predicted a buyer, the false positives that fooled keyword matching, a phrase list you can paste into searches, and the rubric the labels were made with.

**Disclosure:** I build [Needle](https://useneedle.net), and this dataset was built to evaluate its lead scoring. The labels were made by an AI model, not by humans (details and the agreement check are [below](#how-the-dataset-was-built)). If you want the intent categories explained conceptually first, read [5 buying intent signals you're probably missing](https://useneedle.net/blog/5-buying-intent-signals-missed). This post is the data behind them.

---

## In brief

- **Leads are rare.** 83 of 1,416 keyword-matched posts (5.9%) were leads; 21 (1.5%) were "hot". In the random slice, 19 of 620 (3.1%, 95% CI 2.0-4.7%).
- **The biggest false positive is the keyword itself.** 53% of matches weren't about the product's domain at all. Another 30% of all matches were self-promotion, and not one of the 83 leads was.
- **Asking other people is the strongest signal.** "What do you use for", "anyone tried", "how do I", "suggestions", "cheaper/budget" and "first time / new to" were 3-9x more common in leads than in off-topic matches. "Looking for" was only 1.5x, because ads use it too.
- **Half of the worth-replying posts were how-to questions, not shopping requests.** If you only search "recommend" and "alternative to", you miss them. Switching posts were rare: 7 of 83 leads.
- **A simple phrase filter triples precision but misses most leads.** Seek phrase + question mark + not promo lifted the lead rate from 5.9% to 17.2%, but caught only 41% of leads.

---

## How many keyword matches are real buying intent?

Each post got one of four tiers, judged against what the matched product actually sells, not against keyword overlap:

| Tier | Meaning | Posts | Share of 1,416 | Random slice (620) |
| --- | --- | ---: | ---: | ---: |
| Hot | Actively looking for, comparing or about to pick this kind of product | 21 | 1.5% | 3 (0.5%) |
| Worth replying | A concrete problem or how-to question the product solves | 62 | 4.4% | 16 (2.6%) |
| Monitor | On topic, but no need the product can serve | 582 | 41.1% | 262 (42.3%) |
| Irrelevant | Not about the product's domain at all | 751 | 53.0% | 339 (54.7%) |
| **Leads (hot + worth)** | | **83** | **5.9%** | **19 (3.1%)** |

Why two columns? The 1,416 rows were sampled in layers. 620 were drawn uniformly at random; the rest were deliberately enriched with posts that looked like asks or comparisons, and posts a model scored highly, so there would be enough leads to study. The random column is the better guess at what raw keyword matching returns. The full column has more leads to learn from. The hot count in the random slice is 3, which is too small to quote as anything firmer than "rare".

These shares describe this sample of keyword-matched posts. They aren't a measure of how much buying intent exists on the internet, and a sharper keyword list would do better than the average list here.

---

## Which phrases actually signal buying intent?

I ran a set of phrase patterns over every post (title plus up to about 1,400 characters of text) and counted how often each appeared in leads versus in irrelevant posts. "Lift" is the share among leads divided by the share among irrelevant posts. "Lead rate" is how many of all posts containing the phrase were leads, against a 5.9% baseline.

| Pattern | Leads with it (of 83) | Irrelevant with it (of 751) | Lift | Lead rate when present | Random slice: leads with it (of 19) |
| --- | ---: | ---: | ---: | ---: | ---: |
| "what do you use / recommend / suggest" | 3 | 0 | n/a | 3 of 7 | 0 |
| "anyone use / tried / using" | 3 | 3 | 9.0x | 3 of 11 | 0 |
| "how do I / how can I" | 7 | 14 | 4.5x | 7 of 32 (22%) | 2 |
| "cheaper / affordable / budget" | 9 | 18 | 4.5x | 9 of 57 (16%) | 2 |
| "suggest / suggestions" | 15 | 37 | 3.7x | 15 of 77 (19%) | 4 |
| "hire / agency / expert / consultant" | 6 | 17 | 3.2x | 6 of 59 (10%) | 0 |
| "first time / new to / beginner" | 10 | 29 | 3.1x | 10 of 64 (16%) | 2 |
| "recommend / recommendation" | 17 | 54 | 2.8x | 17 of 110 (15%) | 4 |
| "switching from / moving off" | 2 | 7 | 2.6x | 2 of 16 | 1 |
| Title ends with "?" | 32 | 124 | 2.3x | 32 of 266 (12%) | 9 |
| "I / we need, want, am trying" | 19 | 95 | 1.8x | 19 of 165 (12%) | 3 |
| Any price or budget word ($, price, cost, per month, subscription) | 25 | 129 | 1.8x | 25 of 282 (9%) | 7 |
| "?" anywhere in the post | 60 | 341 | 1.6x | 60 of 680 (9%) | 13 |
| "looking for" | 27 | 168 | 1.5x | 27 of 328 (8%) | 4 |
| Frustration words (struggling, tired of, annoying...) | 11 | 76 | 1.3x | 11 of 130 (8%) | 2 |
| "alternative / alternatives" | 2 | 14 | 1.3x | 2 of 26 | 0 |

Read the small rows with care. Anything with fewer than about 10 leads behind it ("what do you use", "anyone tried", "switching", "alternative") is a hint, not a finding. The random-slice column is there so you can see which patterns hold up outside the enriched rows: "recommend", "suggest" and "title ends with ?" do; the tiny ones can't be judged.

Three things surprised me:

1. **"Looking for" is a weak signal on its own.** It shows up in a third of leads, but also in more than a fifth of off-topic posts. Ads, listings and community recruitment posts open with "Looking for a...?" all the time.
2. **Frustration barely separates anything.** 13% of leads had frustration words, and so did 10% of irrelevant posts. People complain about everything. Frustration only matters when it's about the job your product does.
3. **"Alternative" was not a strong signal in this sample.** Only 2 leads used it. That's partly small numbers and partly that "alternative" turns up in listicles and "free alternatives to X" roundups. It's still worth searching (see [how to mine switching intent](https://useneedle.net/blog/alternatives-and-switching-intent-mining)), just don't assume every hit is a buyer.

Post length didn't separate leads from non-leads (median about 236 words for leads, 242 for irrelevant), though text was capped at about 1,400 characters, so long posts were cut.

---

## What does a buying-intent post look like? Paraphrased examples

These are **paraphrased composites**, not quotes. Each one is generalised from the pattern of several labelled posts, with categories replaced by placeholders, so no real post or author can be identified.

| Tier | Paraphrased composite | Why it got that tier |
| --- | --- | --- |
| Hot | "Opening a small online store. Any recommendations for a [category] tool that won't cost a fortune?" | Explicit seek + budget + fits the product |
| Hot | "Comparing [tool A] and [tool B] for [job]. Which did you go with and why?" | Choosing between products of this kind |
| Hot | "Need an [service] agency for our store, has anyone hired one they'd recommend?" | Seeking a service the brand sells |
| Worth | "First time doing [task]. How do I set it up without doing it by hand every week?" | Concrete problem the product solves, no shopping ask |
| Worth | "Our [metric] dropped after [change]. Where would you start?" | Domain how-to where a helpful reply could mention a tool |
| Monitor | "We just launched [product], a [category] tool. Feedback welcome!" | Author promoting their own product |
| Monitor | "Looking for a [service] in [city]" when the product serves a different city | Right topic, wrong place |
| Monitor | "Looking for a new group? Join our [community], friendly members, active mods" | Recruitment ad that happens to say "looking for" |
| Irrelevant | A post using the keyword in a different sense entirely | Keyword collision |

---

## What are the most common false positives?

Most of the noise falls into a few buckets.

**Across all 1,416 matches:**

| False-positive pattern | Posts with it | Leads among them | Lead rate |
| --- | ---: | ---: | ---: |
| Not about the product's domain (labelled irrelevant) | 751 | 0 | 0% |
| Author promoting own product, service or project (labelled promo) | 420 | 0 | 0% |
| Contains a URL | 267 | 1 | 0.4% |
| "I / we built, made, launched, shipped" | 88 | 0 | 0% |
| News words (announced, report, raises, acquired) | 83 | 5 | 6% |
| "vs / versus / comparison" | 70 | 2 | 3% |
| Tutorial words (guide, tips, here's how) | 43 | 2 | 5% |

The "vs" row is worth a second look: you'd expect comparisons to be buyers, but most were reviews, listicles and opinion pieces, not people choosing.

**Inside the on-topic but not-a-buyer group (monitor, 582 posts)**, I bucketed the labeller's short reason for each post. These buckets are approximate, since they come from keyword-matching a 12-word note:

| Why an on-topic post wasn't a lead | Posts | Share of monitor |
| --- | ---: | ---: |
| Author promoting their own product or service | 227 | 39% |
| Needs something the product doesn't offer, or wrong location or segment | 123 | 21% |
| Opinion, discussion or meme | 66 | 11% |
| News or link share | 34 | 6% |
| Job post or job seeking | 31 | 5% |
| Tutorial, showcase or "how I did it" | 26 | 4% |
| Other or unclassified | 75 | 13% |

The second row is the one keyword tools can't fix: someone really is shopping, just not for what you sell. A request for a local service in another city, or a tool that does a neighbouring job, reads like a lead and isn't one.

---

## What kinds of intent do real leads show?

A subset of 250 posts (all 83 leads plus 167 monitor posts) also got an intent type: looking for a tool, a how-to question, switching, pain without an ask, or none.

| Intent type | Hot (21) | Worth (62) | All leads (83) | Monitor sample (167) |
| --- | ---: | ---: | ---: | ---: |
| Looking for a tool or service | 17 | 16 | 33 (40%) | 11 |
| How-to or advice question | 1 | 32 | 33 (40%) | 33 |
| Switching or comparing | 2 | 5 | 7 (8%) | 5 |
| Pain, no ask | 0 | 3 | 3 (4%) | 0 |
| None | 1 | 6 | 7 (8%) | 118 |

Two takeaways. Hot posts are almost all explicit seeks (17 of 21). And half of the worth-replying posts (32 of 62) were plain how-to questions, so a search built only on "recommend" and "alternative to" will miss them. Note the 11 monitor posts that were looking for a tool: they were shopping for something the matched product doesn't sell. Intent without fit isn't a lead.

These intent types were mostly from a single labeller (181 of 250 rows), with no agreement score, so treat this table as softer than the tier counts.

---

## Does a phrase filter work?

I tried the obvious shortcut: keep posts that contain any seek phrase (looking for, recommend, suggest, alternative, anyone tried, best X for, switching from, need a tool), contain a question mark, and aren't self-promotion.

| Filter | Sample | Posts kept | Leads kept | Lead rate | Share of all leads caught |
| --- | --- | ---: | ---: | ---: | ---: |
| No filter | All 1,416 | 1,416 | 83 | 5.9% | 100% |
| Any seek phrase | All | 523 | 50 | 9.6% | 60% |
| Seek phrase, not promo | All | 376 | 50 | 13.3% | 60% |
| Seek phrase + "?" + not promo | All | 198 | 34 | 17.2% | 41% |
| No filter | Random 620 | 620 | 19 | 3.1% | 100% |
| Seek phrase + "?" + not promo | Random 620 | 44 | 5 | 11.4% | 26% |

The filter roughly triples the share of leads in what's left, which is a real saving if you read everything by hand. But it drops more than half of the leads, mostly the how-to questions and the posts that describe a need without the magic words. Even after filtering, more than 4 in 5 posts you read still aren't leads. Phrase searches are a good start; they don't replace reading for fit.

---

## Intent phrase list: paste these into your searches

Built from the patterns above, strongest first. Replace `[job]`, `[category]` and `[competitor]` with your own words. The quotes and `site:` work in Google; in a community's own search box, paste the quoted phrase alone.

**Strongest asks (highest lift, small samples):**

```text
"what do you use for" [job]
"anyone tried" [category]
"how do I" [job]
"any suggestions" [category]
"any recommendations" [category]
"recommend a" [category]
```

**Budget and first-timer language:**

```text
"cheaper" [competitor]
"affordable" [category]
"on a budget" [category]
"first time" [job]
"new to" [category] "?"
"beginner" [category] "what"
```

**Services (if you sell expertise, not software):**

```text
"looking for an agency" [service]
"hire" [service] "recommend"
"looking for an expert" [job]
```

**Switching (rare, but high value when it fits):**

```text
"switching from" [competitor]
"moving off" [competitor]
"alternative to" [competitor] "?"
```

**Exclusions** (add to Google queries; they strip out the launch, self-promotion and hiring posts that made up much of the noise in this data):

```text
-"I built" -"we launched" -"just launched" -"check out my" -"hiring" -"giveaway"
```

To limit a query to one community: `site:news.ycombinator.com "what do you use for" [job]`, or swap in any forum's domain. If you want a fuller walkthrough of running this across several communities, see [multi-platform search](https://useneedle.net/guides/multi-platform-search). For why plain keyword alerts drown in the 53% problem, see [keyword alerts vs search](https://useneedle.net/comparisons/keyword-alerts-vs-search-f5bot-graduate).

---

## A buying intent scoring rubric you can use

This is the labelling guide the dataset was built with, rewritten as a checklist. The order matters: decide fit first, then intent.

| Tier | Test | Action |
| --- | --- | --- |
| **Hot** | Explicit seek, compare or choose signal ("any recommendations", "which do you use", "switching from X") **and** it's for the kind of thing you sell, in a place and segment you serve | Reply today, answer the actual question first, disclose who you are if you mention your product |
| **Worth replying** | A concrete problem or how-to question your product solves, or frustration with a competitor without asking for alternatives | Reply with a genuinely useful answer; mention the product only if it's the natural answer |
| **Watch** | On your topic, but no need you can serve: news, opinion, tutorial, job post, someone promoting their own product, or a need in a market you don't serve | Don't pitch. Read for language and objections |
| **Ignore** | Not about your domain: a keyword collision or different meaning | Skip, and add a negative keyword |

Five rules from the guide that did most of the work:

1. **Be strict with hot.** It needs an explicit seek signal *and* a fit. Seeking something you don't sell is "watch" at best.
2. **A seller fishing for feedback is never a lead.** "I built X, thoughts?" got zero leads in 420 promo posts.
3. **Location and audience matter.** A request for a service in another city is not a lead for yours.
4. **Judge what's shown.** If the post is cut off, don't fill in intent that isn't there.
5. **When torn between two tiers, pick the lower one.** It keeps your hot list honest.

For what to actually say once you've found a hot or worth post, the [community reply templates](https://useneedle.net/guides/community-reply-templates) cover disclosure and tone. For spotting budget signals specifically, see [willingness-to-pay signals in public conversations](https://useneedle.net/blog/willingness-to-pay-signals-public-conversations).

---

## How the dataset was built

- **Posts:** 1,416 public posts captured October 4-17, 2026, each matched to the keyword list of one of 202 products (a mix of B2B software, agencies, consumer goods and local services). Duplicates by product and text were removed, with at most 2-3 posts per product per sampling layer.
- **Sampling:** 620 posts uniformly random; 516 enriched with posts matching ask or comparison patterns; 200 from the top of a model's score; 80 from non-Reddit platforms. Because the enriched layers were picked partly on ask phrases, phrase lift in the full sample is a little flattered, which is why the random-slice column is shown.
- **Labels:** an AI model (Claude) labelled every post against a written guide, from up to 1,500 characters of text, blind to every model score. A second, independent AI labeller re-labelled a random 30% (424 posts). They agreed on 93.4% of tiers (Cohen's kappa 0.88, a chance-corrected agreement score where 1.0 is perfect) and on 99.5% of lead versus non-lead calls (kappa 0.94). The three lead-relevant disagreements were settled by hand; monitor-versus-irrelevant disagreements took the lower tier.
- **Promo** was a separate yes/no flag: is the author announcing or promoting their own product, project or service.

**Platforms in the sample.** The sample is heavily skewed, so this table shows what was in the dataset, not which platform has more buyers:

| Platform | Posts | Leads | Promo flagged |
| --- | ---: | ---: | ---: |
| Reddit | 1,234 | 82 | 338 |
| Bluesky | 166 | 1 | 70 |
| Hacker News | 13 | 0 | 12 |
| Stack Overflow | 3 | 0 | 0 |

With 13 Hacker News posts and 3 Stack Overflow posts, there's nothing to conclude about those platforms. Bluesky's 1 lead in 166 is worth a footnote, not a verdict.

---

## Limitations

- **AI labels, not human ones.** Two independent AI labellers agreeing at kappa 0.88 is a good consistency check, but it doesn't prove either matches what a human buyer intended. Small differences (a few posts) shouldn't be over-read.
- **Small numbers at the top.** 83 leads and 21 hot posts are enough for rough patterns, not precise rates. Several phrase rows rest on fewer than 10 leads.
- **Keyword-matched, not the whole internet.** Every post had already matched a product's keyword list. A better keyword list would produce a higher lead rate; a broad one, lower.
- **Platform skew.** 87% of posts came from one platform, so the patterns mostly describe that platform's writing style.
- **Two weeks of posts, one period.** October 4-17, 2026. Seasonal or news-driven topics could shift the mix.
- **Truncated text.** Posts were capped at about 1,400 characters, so signals late in long posts were missed and length can't be judged.
- **English only, mostly.** 11 posts were flagged as mostly non-English; the phrase list is English.

If you're choosing a tool to do this filtering for you rather than by hand, I compared options in [best buyer intent and social lead discovery tools](https://useneedle.net/blog/best-buyer-intent-social-lead-discovery-tools-2026). Whatever you use, check its hot list against the rubric above: if more than a few of its "hot" posts are self-promotion or the wrong city, the scoring is keyword matching with extra steps.

---

## Evidence notes

- **Primary source:** a private labelled dataset of 1,416 public posts (captured October 4-17, 2026), with tier labels, promo and non-English flags, and intent-type labels for a 250-post subset. All counts in this post were computed from it on November 21, 2026 with a script that re-reads the raw label files. The dataset isn't published because it includes the keyword lists of the products it was matched to.
- **Labelling guide:** the four-tier rubric and the five rules in the rubric section are the guide the labels were made with, lightly reworded.
- **Confidence intervals** are 95% Wilson intervals: lead share 5.9% (4.8-7.2%) in the full sample, 3.1% (2.0-4.7%) in the random slice; self-promotion 29.7% (27.3-32.1%).
- **Examples** in the examples table are paraphrased composites, not quotes, and contain no product, author or community names from the data.
- No external statistics are used in this post.

## FAQ

### What percentage of keyword-matched social posts are real buying intent?

In a labelled set of 1,416 public posts that matched product keyword lists, 83 (5.9%, about 1 in 17) were worth a reply and 21 (1.5%) showed someone actively looking to buy. In the uniformly random part of that sample, 19 of 620 posts (3.1%, about 1 in 33) were leads. Most keyword matches were off-topic or self-promotion.

### What are the strongest buying intent signals in a social post?

In this dataset, phrases asking other people what they use ("what do you use for", "anyone tried"), "how do I" questions, recommendation and suggestion requests, price words like "cheaper" or "budget", and first-timer language ("new to", "first time") were 3 to 9 times more common in leads than in off-topic matches. A title that ends in a question mark was 2.3 times more common. None of these is enough alone: the post also has to fit what you sell.

### Is "looking for" a good buying intent keyword?

It is weaker than most people expect. "Looking for" appeared in 33% of leads but also in 22% of off-topic posts, a lift of only 1.5x, because ads and community recruitment posts use it constantly ("Looking for a new group to join? Welcome to..."). Pair it with a category word and a question mark, and exclude launch and promo language.

### How do I score a post for purchase intent?

Use four tiers. Hot: an explicit seek, compare or choose signal for the kind of product you sell. Worth replying: a concrete problem or how-to question your product solves, without an explicit shopping request. Watch: on your topic but no need you can serve, including news, job posts and self-promotion. Ignore: not about your domain. When torn between two tiers, pick the lower one.

### Were these posts labelled by humans?

No. The tiers were assigned by an AI model following a written labelling guide, blind to any scoring. A second, independent AI labeller re-labelled a random 30% (424 posts); they agreed on 93.4% of tiers (Cohen's kappa 0.88) and on 99.5% of lead versus non-lead calls (kappa 0.94). The three lead-relevant disagreements were settled by hand.

---

From Needle (https://useneedle.net). HTML version: https://useneedle.net/blog/buying-intent-posts-1416-labelled-examples
