earlyink
203 first flagged21 Oct 2026

Stat Trace: find where a number in an AI answer came from

A browser extension that traces statistics in AI answers to their earliest source. Writers and analysts pay.

Competitive · Fake stats now get cited by LLMs

Written for: Solo builder

Next step

Prototype the trace on 50 statistics taken from real AI answers about marketing and SaaS, publish how many trace back to one source, and share the write-up and extension on HN and r/content_marketing.

Why now

People now copy numbers from AI answers straight into posts, decks and client reports. Checking provenance by hand means hunting for the earliest page that stated the figure and asking whether anyone measured it independently. Search APIs and small models make that hunt cheap to automate.

What became possible

An HN discussion shows that publishing invented statistics gets them cited by LLMs. Toolflation shows seller agents spontaneously gaming listings that buyer agents read. Brand-preference research finds models converge on shared recommendations formed in pretraining. Together they suggest AI answers are an easy target and that a single planted source can spread.

Three ways in

  1. Stat Trace extension

    Highlights every statistic in an AI answer, searches for the earliest page that stated it, and counts the independent sources that report it. It flags single-source and circular figures.

    Customer
    Content marketers and freelance writers
    Model
    Free for 20 traces/month. $6/month unlimited
  2. Draft check

    Paste a finished article and get a provenance table for every number in it before publishing.

    Customer
    Newsletter writers and small editorial teams
    Model
    $15/month team plan
  3. Planted-stat feed

    A public list of statistics found to come from a single invented source, which also serves as marketing.

    Customer
    Journalists
    Model
    Free

Competition

Competitive

Checking claims in AI answers is already crowded. At least eight Chrome extensions (ScrutinyAI, Fact It, Theia, Verifi, Verol, Verity and others) give verdicts on ChatGPT, Claude and Gemini output.

Full competition notes (12 competitors)

Checking claims in AI answers is already crowded. At least eight Chrome extensions (ScrutinyAI, Fact It, Theia, Verifi, Verol, Verity and others) give verdicts on ChatGPT, Claude and Gemini output. Pre-publish checkers for marketers also exist: GPTZero Source Finder, Originality.ai, LongShot and Fractal. A plain "is this number supported by a source" product would be a commodity, and GPTZero and Originality already have distribution with this exact buyer. I found nothing that does the specific job in this idea. That job is finding the earliest page where a figure appeared, telling independent sources apart from pages that copied each other, and flagging stats that come from one source or that cite each other in a loop. Existing tools answer whether some page supports the claim. That is the weak spot for zombie stats, because a fake figure usually has dozens of supporting pages that all lead back to one vendor blog. A new entrant could win by being the provenance tool rather than a fact-checker. It would show the citation chain, the first date the number appeared and an independence count, plus a public feed of debunked stats as a marketing and data advantage. The risks are cost and accuracy. Finding the earliest appearance means searching for the figure in many forms and using archive dates, which is expensive and often uncertain. Incumbents could also add a "trace to origin" feature if the idea gets traction. The paid market for writers is real but price-sensitive.

  • GPTZero Source Finder

    Finds the checkable claims in pasted text, matches them to web and academic sources, marks each claim as supported or contradicted, and exports citations. It has a large existing user base. It finds a source but does not look for the earliest one or count how many sources are independent.

  • Originality.ai Fact & AI Hallucination Checker

    Flags statements that may be false, explains why and links sources. It is sold to content marketers before publishing, which overlaps directly with the Draft check idea.

  • LongShot Fact Check & Cite

    Takes pasted text or a document, detects the claims, labels whether they are accurate, links supporting sources and can swap in corrections. It is aimed at SEO writers.

  • Fractal Fact Check Agent

    An editing agent that picks out stats, dates and quotes that need checking and verifies them against a source index.

  • ScrutinyAI

    Chrome extension that pulls the claims out of ChatGPT answers and gives each a web-backed verdict: source confirmed, externally supported, contradicted, or differing view.

  • Fact It

    New extension that underlines factual claims in ChatGPT, Claude and Gemini answers while they stream, with a supported, partial or contradicted signal.

  • Theia

    Extension that fact-checks ChatGPT, Claude and Gemini answers and links the sources behind each verdict.

  • Verifi AI

    Extension that fact-checks AI answers in real time with sources and confidence scores. It is tiny, with about nine users.

  • Verol

    Marks risky claims inline in ChatGPT, Claude, Gemini and Grok answers with color-coded indicators.

  • Verity

    Gives each claim a badge (verified, wrong, partially verified or unverifiable) with a link to its source.

  • Citely Source Finder

    Matches pasted text to its original source, mostly in academic databases. It is the closest to tracing a claim back to where it started, but it is built for academic citations, not web statistics.

  • StoryTracer

    Chrome extension that follows a news page's outbound links to guess where the original reporting came from. It traces whole stories, not individual numbers.

Checked 2026-10-21

Risks

  • Finding the earliest source is hard when pages lack dates or are rewritten
  • Search API costs eat a low-priced subscription
  • Answer engines may add inline provenance themselves
  • Market pull is soft, since most people do not check numbers

Strong team fit; market pull is the weak point (4).

tech 7 · pull 4 · space 6 · fit 8 · build 7

Exposed

Companies that could benefit or be threatened. News and context, not investment advice.

Momentum of its main topic

180 developments this week +41 vs last week
22 Sep: 0 developments23 Sep: 0 developments24 Sep: 0 developments25 Sep: 0 developments26 Sep: 0 developments27 Sep: 0 developments28 Sep: 0 developments29 Sep: 0 developments30 Sep: 0 developments1 Oct: 0 developments2 Oct: 0 developments3 Oct: 0 developments4 Oct: 0 developments5 Oct: 0 developments6 Oct: 0 developments7 Oct: 0 developments8 Oct: 32 developments9 Oct: 26 developments10 Oct: 13 developments11 Oct: 11 developments12 Oct: 14 developments13 Oct: 19 developments14 Oct: 24 developments15 Oct: 35 developments16 Oct: 31 developments17 Oct: 15 developments18 Oct: 14 developments19 Oct: 23 developments20 Oct: 26 developments21 Oct: 36 developments
Stage: Coverage

Sources

  1. Publish fake stats, get cited in LLMsHacker News
  2. Toolflation: Seller Agents Distort Tool Markets Through Spontaneous Generative Engine OptimizationarXiv
  3. How Language Models Learn Taste: Brand Preference in LLMsarXiv

Status history

  • 2026-10-21 Flagged as buildable