Research Q&A with verifiable sources

Experiment method — not executed

Status: not executed. No answer engine was run for this guide, so it publishes no measured answer, citation check, latency, score or ranking. What follows is the reproducible plan and the conditions that are missing.

The intended test gives every cited-answer tool the same fixed set of questions and checks whether the sources it cites actually contain the claim, not whether the answer sounds confident. Nothing was executed in this round — no account, API key or entitlement was used — so the plan below has no results and this guide reports documented citation behaviour only.

Method (planned, not yet executed)

• Test target: whether a tool's answer cites a source that really contains the claim. Two questions have an answer that appears only in material published after most models' training data (the freshness probe), and one is unanswerable from the sources on offer (the honesty probe).
• Fixed input: every tool would receive the identical five questions from one file (docs/tests/research-inputs.md, still to be created), so the wording is the same for all of them.
• Task: ask each tool the same five questions. A tool passes the freshness probe only if it cites a source that actually contains the fact, and passes the honesty probe only if it declines.
• Citation check: for every citation URL a tool returns, open the URL, read it, and record whether it loads and whether the page actually contains the claim it is attached to. A URL that does not load, or a page that does not state the claim, is recorded as not supporting it.
• Recorded per tool: product and plan (free tier or paid), account type, and the exact model or mode selected; the full answer text, unedited, with every citation URL; token or call usage and the metered cost if the tool bills per use; and the date, time and region of the run.
• Environment required before any run: a reachable, signed-in web answer engine and, where the tool bills per use, a funded key. None of these was available to this project in this round.

Missing conditions — what is unconfirmed, and what this does not show

• No run was executed this round, so no answer, citation check, latency or cost is reported for any tool below.
• ChatGPT: an account is required to use the product; a free tier exists, and a paid plan is needed only for the newer models. Its pricing page answered HTTP 200 on 2026-10-10 but renders prices in JavaScript, so no price was read.
• Perplexity: an account is required; a free tier exists and Pro is the paid plan. Its Pro page answered HTTP 403 to this project's fetcher on 2026-10-10, so no plan price was read.
• You.com: the API needs a key and a signed-in account; new keys come with free credits, so a paid plan is not the precondition.
• Genspark: an account is required and a free plan exists. The public help-centre pricing page answered HTTP 200 on 2026-10-13, and its plan prices are recorded on the product page.
• Microsoft Copilot: a Microsoft account is required; core Copilot features are free and the paid plan is optional. The entitlement actually needed for a cited-answer run was not confirmed this round.
• Exa: the API needs a key; new developers receive free credits, so a paid plan is not the precondition.
• Where the plan a tool needs in order to run is not confirmed, it is recorded as unconfirmed rather than assumed to be paid. A free tier being available does not mean a free-tier run was performed — none was.
• This is the plan for a method, not a benchmark: it cannot say which engine cites more accurately, and it does not rank, score or compare the products listed in this guide.

Cost. Nothing was run, so there is no measured cost and no metered amount was recorded. No account or paid plan was purchased for this theme; only the vendor pages named below were read.

Public method and copies

The first entry is the full method document published on this site; the second is the index of every published copy. There is no run record here, because no run exists.

• docs/tests/source-backed-research-qa.md
• docs/tests/

What was not executed

ChatGPT · Perplexity · You.com · Genspark · Microsoft Copilot · Exa

This round executed none of these products, so this page reports no measured result for them. The condition each one needs is stated above; where a condition was not confirmed it is recorded as unconfirmed, not assumed to be paid. Their pages and comparisons report documented behaviour with a source and a check date.