Paste a product URL into directree and, in about 30 seconds, you can have a structured software listing. The important part is not the speed. It is that we show where the information came from, what the system inferred, and what still needs a founder's correction.
This is a build-in-public look at that flow. No magic black box. No claim that one URL can tell us everything about a product.
The short version: one URL is a starting point, not a verdict
A software website already contains more useful public information than most directory forms ask a founder to type in from scratch: its page title, description, visible feature copy, pricing language, social profiles, screenshots, and sometimes structured data intended for search engines.
directree reads that public page, turns the verifiable parts into Observed facts, then asks an AI step to produce clearly labelled AI-inferred context. Finally, a verified owner can claim the listing and replace or add information as Founder-edited.
Those are three different claims. We keep them separate on purpose. If the crawl found a free-trial button, we can say we observed one. If the page suggests the product is aimed at small marketing teams, that is interpretation. If the founder corrects the positioning after verifying the domain, that is their accountable statement.
For the full philosophy behind those labels, read the directree honesty model. This post is about the machinery underneath it.
Step 1: We fetch the public page and collect only what it actually says
The first job is a crawl of the URL that was pasted in. We look for the ordinary public signals a visitor can see or a website exposes in its HTML:
- Page title, meta description, and Open Graph details
- A logo, favicon, and social profile links
- Visible page copy, including headings and body text
- Pricing, free-plan, and free-trial language when it is publicly visible
- Named product integrations where the page actually mentions them
- Structured data such as JSON-LD, when the site provides it
That becomes the factual layer of the listing. We call it Observed because it is tied to a public source, not because it is timeless or perfect. A company can change its price tomorrow. A crawler can miss a JavaScript-heavy section. A page can contain stale copy. Observed means “we saw this on the public site,” not “this is forever true.”
There is a useful restraint here too. We deliberately do not scrape a whole website, visit every pricing tier, or pretend we have tested the product. The current enrichment pass begins with the URL you give us. If a fact is not available there, it is better to leave it unclear than fill the gap with something plausible-sounding.
That restraint is why directree can create a useful first listing quickly without quietly turning an assumption into a fact.
Step 2: AI adds context, but never gets to impersonate evidence
Raw site data helps, but it does not answer every question a buyer has. A feature page might tell you what a tool does without plainly saying who it is best for, what trade-off it makes, or what category it belongs in.
That is where the AI-inference step comes in. It receives the observed data and the visible copy extracted from the page. It can propose a neutral summary, likely audience, key features, likely competitors, tags, and a best-fit category. It can also suggest a strength and a limitation.
But it is instructed to stay grounded in the page text, avoid invented pricing, avoid fake numeric scores, and hedge where public information is thin. An inferred weakness is not framed as a finding from a product test. It is explicitly marked as AI-inferred and may be outdated. A founder can correct it.
This distinction matters more than it may sound. A directory can make a polished AI summary look like an editorial review with a few styling choices. That is misleading by layout. We would rather show a helpful interpretation and say exactly what it is.
Our post on why directree refuses fake precision scores comes from the same principle. A clean decimal or a confident paragraph can both create the impression of certainty. Neither deserves it unless the underlying method supports it.
Step 3: You choose a quick listing or a claimed one
After enrichment, the submit flow gives founders two free paths.
Quick AI listing is for a founder who wants a listing live without spending time filling out fields. They get a read-only preview, then publish the structured listing as-is. The AI-derived parts remain labelled as AI-inferred. It is useful for getting a public starting point, but it does not make the founder the source of the listing.
Claim & build it yourself starts with the same enrichment, then moves into ownership verification. Once the founder verifies that they represent the domain, they can edit their listing. Those changes are visibly marked Founder-edited rather than blended into crawl output.
That is not just a product-flow distinction. It is an accountability distinction. Creating a listing and owning a listing are different things. A directory should not let somebody who merely pasted a URL overwrite a company’s public facts.
If you are deciding whether that backlink is worth bothering with, our guide to do-follow versus no-follow backlinks explains the mechanics and the limits. It is one relevant link, not a promise of instant rankings.
What the 30-second flow does not do
Speed makes a good demo. Limits make a trustworthy product. Here are the things the first enrichment pass does not claim to do:
- It does not audit the entire product. We are reading a public web page, not running a full technical, security, or usability review.
- It does not guarantee every field is current. Websites change. The listing records what was publicly available when it was processed, and founders can correct it after verification.
- It does not turn a guess into a fact. Best-for, not-for, strengths, weaknesses, and summaries are clearly labelled AI-inferred.
- It does not use a composite score to rank products. We do not think “8.7/10” is honest when the method is not auditable.
- It does not give a founder control before they prove ownership. A submitter can create a listing, but verification is the gate for Founder-edited claims.
Those constraints cost us some apparent neatness. A listing may have an unknown field. An AI insight may be hedged. A founder may need to take two minutes to verify and improve the page. Good. The alternative is a directory that looks complete by concealing uncertainty.
Why we built it this way
We have built and marketed businesses since 2008, and one lesson has repeated across five businesses: a good product can still stay invisible if nobody can find a clear, credible explanation of it.
The usual answer is a long submission form that founders postpone, or a thin auto-generated page that buyers cannot trust. We wanted the useful middle: reduce the work of getting listed, while making the data provenance visible enough that a buyer can judge it properly.
A URL is therefore not an application form. It is a first piece of evidence.
The crawler gives the listing a factual base. AI makes the first pass easier to browse. Founder verification gives the product owner a visible way to improve it. Community signals and directree surveys can then help people discover tools without pretending that a decimal score settles the question.
That is the product we are trying to build: not a directory that knows everything, but one that does not hide what it knows, what it infers, or who last edited the claim.
FAQ
How long does a directree listing take to create?
The initial public-page enrichment and structured listing are designed to take about 30 seconds. Actual timing can vary with the site being fetched and the amount of information available.
What information is marked Observed?
Observed fields come from public signals on the supplied site, such as visible copy, metadata, pricing language, integrations mentioned on the page, social links, and structured data where available.
Can founders change AI-inferred information?
Yes. After a founder verifies ownership of the domain, they can edit their listing. Their corrections appear as Founder-edited, so readers can see that a verified representative supplied them.
Why not crawl every page on a product website?
A fast first pass needs clear limits. Crawling every subpage can create more stale or contradictory data and makes it harder to explain exactly where a claim came from. We would rather start from the submitted URL, label uncertainty, and let a verified founder improve the record.
Does a directree listing guarantee better search rankings?
No. A claimed listing can provide a relevant do-follow backlink through the verification badge, but no single link guarantees rankings. Visibility comes from a wider body of useful content, technical basics, relevant links, and time.