The ChatGPT app and its API cite different sources.

    We asked the ChatGPT app and its API the same five US buying questions, three times each, at the same time. Only 7 of 55 cited sites appeared on both sides. Honeyb captures the app, in a real browser session.

    See the answers

    We issue keys within one working day. Tell us your engines and volume.

    Of cited sites shared, 7 of 55
    13%
    US buying questions
    5
    Runs on each side
    3

    FIG 01

    The answer buyers see cites reviews and guides. Its API almost never does.

    ChatGPT app, as Honeyb captures it The same model through its API

    Reviews, guides and publishers among the sources cited

    24 of 37
    1 of 109
    Five buying questions, three runs on each side, asked in the US at the same time. A review, guide or publisher is a cited site that belongs to none of the brands the answers named.

    Side by side

    The same question, answered twice.

    Only quickbooks.intuit.com appeared on both lists. The rest of the evidence behind each answer was different.

    Asked of bothWhat accounting software should a US startup use?

    ChatGPT app

    3 web searches, a 5,410-character answer

    Brands named

    • QuickBooks
    • Xero
    • Puzzle (only this side)
    • Zoho Books (only this side)
    • Wave (only this side)
    • NetSuite
    • Mercury (only this side)
    • Gusto (only this side)
    • Ramp
    • Carta (only this side)
    • Stripe

    Sources cited

    • dupple.com
    • quickbooks.intuit.comCited by both
    • startupcfo.ai
    • zoho.com

    GPT API with web search

    2 web searches, a 4,812-character answer

    Brands named

    • QuickBooks
    • Xero
    • NetSuite
    • Ramp
    • Stripe
    • Shopify (only this side)

    Sources cited

    • netsuite.com
    • quickbooks.intuit.comCited by both
    • ramp.com
    • xero.com

    Named or cited by one side onlyPlain: named or cited by both

    All 5 questions

    The brands overlap. The sources barely do.

    33 of 60 brands appeared on both sides, and the two put the same brand first on all 5 questions. Only 7 of 55 cited sites appeared on both. The evidence is what a brand can change.

    What accounting software should a US startup use?

    Brands8 of 18

    Sites1 of 11

    Top pickYes, QuickBooks

    What are the best billing platforms for SaaS companies?

    Brands9 of 11

    Sites0 of 20

    Top pickYes, Stripe

    What is the best email marketing platform for a Shopify store?

    Brands4 of 8

    Sites4 of 10

    Top pickYes, Klaviyo

    What is the best project management software for a marketing agency?

    Brands6 of 17

    Sites1 of 5

    Top pickYes, Teamwork

    Which CRM is best for a 10-person sales team?

    Brands6 of 6

    Sites1 of 9

    Top pickYes, Pipedrive

    All 5 questions

    33 of 60

    7 of 55

    5 of 5 matched

    Why it matters

    The pages that shape the real answer are guides and reviews.

    In the ChatGPT app, 24 of 37 cited sources were third-party guides and review sites. Through the API, 108 of 109 were the vendors’ own pages, the one kind of page a brand already controls.

    Data from a model API points a brand at its own website and misses the pages the app reads. Honeyb captures ChatGPT and all twelve engines in real browser sessions, so the sources you see are the ones to win.

    How we measured

    Two surfaces, the same questions, at the same time.

    Questions
    5 US buying questions, 3 runs of each on each side, all submitted together.
    ChatGPT app
    ChatGPT app, captured in a real browser session with web search on.
    GPT API with web search
    OpenAI GPT-5.6 model through an API, hosted web search on, US location.
    Vendor site or third party
    A cited page counts as a vendor site when its domain belongs to a brand named in the answers to the same question (for example quickbooks.intuit.com, docs.stripe.com); everything else is third-party (review sites, guides, publishers). Counts are cited sources across all 15 answers per surface.
    Brands
    Brand names were read out of every answer by the same model, so both sides were scored by one reader.
    Run to run
    The same question can name a different brand or cite a different page on the next run, which is why each one ran 3 times on each side. AI answers change daily. That's why you need to track them.

    Honeyb API

    Get the answers your buyers actually see.

    All twelve engines captured in real browser sessions, ChatGPT first, with the brands they name and the sources they cite, on one key.

    ChatGPTClaudeGeminiPerplexity
    See pricing

    We issue keys within one working day. Tell us your engines and volume.