Methodology

How this study classifies a homepage

The text below is the actual instruction given to the model (prompt v4) — not a paraphrase — so this page can't drift from what was really asked.

Instructions given to the model

You are a careful research analyst measuring how much a company's homepage states explicitly versus leaves for a reader to infer. You will be given the cleaned text content of a company's homepage, extracted as markdown (heading levels reflect the page's actual structure). For each dimension below, classify it as one of: - Explicit: the homepage states this directly and unambiguously. - Implied: the homepage does not state this directly, but a careful reader could reasonably infer it from context, adjacent claims, or descriptions of the product. - Missing: the homepage gives no basis, direct or inferable, to answer this. Do not use outside knowledge about the company, even if you recognize it -- judge only what this homepage text supports. Be strict about "Explicit": vague marketing language (e.g. "the best tool for modern teams") does not count as explicitly stating a category, customer, or problem unless it actually names one.

The six dimensions

Product category

Does the homepage explicitly state what software/product category this is? Classify as Explicit, Implied, or Missing.

Target customer

Does the homepage explicitly state who the target customer is? Classify as Explicit, Implied, or Missing.

Primary problem

Does the homepage explicitly state the primary problem this product solves? Classify as Explicit, Implied, or Missing.

Core product

Does the homepage explicitly state what the product actually is or does -- concretely, not just marketing language about outcomes? Classify as Explicit, Implied, or Missing.

Differentiation

Does the homepage explicitly communicate a distinctive, concrete reason to choose this product -- a unique capability, a proprietary approach, a specific feature or methodology, or another concrete reason someone would pick it? This counts as Explicit even if no competitor is ever named or compared against -- what matters is a specific, concrete claim, not a competitive comparison. Generic marketing language with no concrete specifics -- "modern," "innovative," "powerful," "best," "purpose-built," "a new species of product" -- does NOT count as Explicit on its own, even if it sounds like a differentiation claim. Classify as Explicit, Implied, or Missing.

Evidence (social proof)

Does the homepage provide concrete evidence of results or adoption? Classify as Explicit only if there is a specific number (e.g. "10,000 customers," a usage stat, an uptime %), a named customer quote or testimonial, or a named case study. A bare row of customer logos with no elaboration -- no names, quotes, or numbers attached -- does NOT count as Explicit; classify that as Implied. Classify as Explicit, Implied, or Missing.

Excluded companies (6)

Companies crawled but left out of the 100-company dataset, and why:

Basecamp	Not part of the official 50-company B2B SaaS sample -- this was the earlier pilot/validation company used while designing the v3/v4 prompt, left in data/companies/ from that phase.
Freshworks	Bot-detection block (Cloudflare 'Just a moment...' challenge, HTTP 403) encountered during the 48->100 expansion -- caught by expand_dataset.py and replaced, left in data/companies/ from the rejected attempt.
Middesk	Crawl captured only nav/footer chrome (476 chars, no hero or body copy) -- likely client-side content that loads after the crawler's load event, not a bot block. Missed by expand_dataset.py's length/phrase heuristics since it wasn't literally a block page; caught manually during the 48->100 expansion and replaced with Baseten.
Salesforce	Bot-detection block (Akamai 'Access Denied') -- homepage never actually rendered, so there was no real content for Claude to analyze.
Sentry	Bot-detection block (challenge page containing 'captcha') encountered during the 48->100 expansion -- caught by expand_dataset.py and replaced with a different candidate, left in data/companies/ from the rejected attempt.
ServiceNow	Bot-detection block (Akamai 'Access Denied') -- same as above.

Scope

This is homepage-only analysis: the model sees only the homepage's cleaned text content, capped at 15,000 characters. It does not read documentation, pricing pages, blog posts, or any other page on the site, and it is not evaluated against AI-search citations, answer-engine visibility, or any off-site source. Those are different, separate metrics that may be added to this Index as future dimensions — this page's per-dimension breakdown updates automatically if they are, with no rebuild required.

Preliminary results. Every classification above is Claude's own reading of the homepage text — not independently verified ground truth. A subset of companies is currently going through manual human review; as reviewed rows are corrected, the numbers on every page of this Index update automatically on the next regeneration.