Skip to content

AI search · Public sources · Citations

Kruqond vs Perplexity: how to audit an AI research answer

A single-task Kruqond and Perplexity comparison, its limitations, and a reusable checklist for checking citations, Reddit comments and video evidence.

By Kruqond · Published

What this comparison can tell you

A long list of citations can look convincing. It does not tell you whether the answer read the comments, checked a video transcript, or found a source that actually answers your question.

This comparison starts with one research task about Obsidian and Notion. Kruqond returned an answer; the Perplexity session stopped at a sign-in prompt. That leaves no fair basis for naming a quality winner, but the Kruqond answer still provides a useful case study in checking AI research.

Disclosure: This article is published by Kruqond, not an independent reviewer. The writing is AI-generated from a saved, automated browser test and source checks. It is a single-task observation, not a general benchmark or a claim that one service is better.

What was compared

Kruqond is an AI search engine and research assistant for public web pages, Reddit posts and comments, YouTube, GitHub, and wider social media including TikTok and Instagram. Access depends on the platform, the source, and the tool selected. Finding a post and retrieving its comments are different jobs.

The test used Kruqond's anonymous default Research & Ask mode, with PRO off and images/videos on. No specialized comment tool was selected. The Perplexity attempt used anonymous default Search without manually selecting a model. Each product received the same prompt once; neither was retried for a more favorable answer.

The question deliberately asks for both a practical recommendation and an explanation of the evidence behind it. A response should make clear whether it found a discussion, read its replies, or only saw a search listing.

What happened in the two sessions

The Perplexity result applies only to that session. It does not establish that Perplexity always requires an account, cannot use Reddit, or cannot retrieve comments. The incomplete pair also cannot support speed, accuracy, coverage, or price comparisons.

  1. Answer observed. Kruqond: Yes, with 12 unique inline source URLs in the saved answer. Perplexity: No answer observed before a sign-in gate.
  2. Reddit comment evidence. Kruqond: The answer explicitly said it had metadata, not comment bodies. Perplexity: Not evaluated.
  3. YouTube content evidence. Kruqond: The answer explicitly said it had metadata and had not watched the videos. Perplexity: Not evaluated.
  4. Quality verdict. Kruqond: Some useful disclosure, plus relevance and claim-qualification issues. Perplexity: Not scored. An access gate is not an answer-quality result.

Check 1: does the citation match the question?

The first named Reddit discussion was Obsidian or OneNote for personal information system? The title and post metadata matched a separately retrieved source. The problem was relevance: the question asked about Notion, not OneNote.

That thread may still offer background on personal knowledge management, but it is not a direct Obsidian-versus-Notion comparison. Treating every citation as equally relevant would hide that difference. A second named thread, Obsidian note taking vs Notion: I don't get it, matched the requested pair more closely.

A useful audit question is simple: would this source still belong in the answer if its title were shown beside the claim? If not, label it background or replace it. Counting links cannot perform that check.

  1. The first named Reddit source compares Obsidian with OneNote, rather than the requested Notion.

    Original Reddit post: Obsidian or OneNote

    What it does not establish: The retrieved title and post metadata establish the topic, not consensus in the comments or a direct comparison with Notion.

  2. The second named Reddit discussion matches the two products in the prompt more closely.

    Original Reddit post: Obsidian and Notion

    What it does not establish: Topic relevance does not establish the arguments in the replies. The saved Kruqond answer did not retrieve those comment bodies.

Check 2: did it read the discussion or find the discussion?

Kruqond's answer said it had Reddit titles, scores, dates, and comment counts, but not comment text. That disclosure matters. A title can establish what a post asks; it cannot establish which arguments appear in its replies or whether commenters agree.

The same distinction applies to TikTok, Instagram, YouTube, and other social platforms. A post with many comments is a lead for further research, not evidence of consensus. Even retrieved comments are only a sample: ranking, pagination, deletions, and access limits can change what is available.

For a comment-based conclusion, ask for the source post URL, the retrieved comment text or a short attributable excerpt, its permalink when available, and the retrieval limits. If the tool only found metadata, keep the answer at that level. This test did not exercise Kruqond's specialized comment tools and does not measure their performance.

Check 3: does a video link support a video claim?

The answer linked a YouTube review titled Obsidian vs Notion – Which Note-Taking App is Better? It also said it had not watched the video. A separate page retrieval yielded the title and minimal page text, not a transcript.

That supports identifying a potentially relevant review, not reporting the reviewer's reasoning. To use a video as substantive evidence, obtain the relevant transcript passage or inspect the content directly. Add a timestamp where possible, and distinguish the creator's claim from a conclusion independently checked elsewhere.

  1. The named review title is relevant to Obsidian and Notion.

    YouTube review named in the answer

    What it does not establish: The saved answer disclosed metadata-only access. Separate retrieval yielded a title and minimal text, not a transcript or video-content verification.

Check 4: follow a product claim to primary documentation

The answer repeated wording from a Zapier comparison that automatic Notion sync requires a premium plan. The source contained that wording, so this was not an invented attribution. It still needed qualification against Notion's own documentation.

Notion's offline help page, checked for this audit, distinguishes three things:

Automatic offline downloading is therefore not interchangeable with every form of syncing or background updating. When a recommendation depends on that distinction, link the primary documentation and preserve the exact condition. A citation can be faithfully quoted and still be too broad for the decision being made.

  1. Members on all plans can download individual pages for offline use in the desktop or mobile apps. Browser offline use is not supported by that feature.
  2. Paid plans automatically download recently visited and favorited pages.
  3. Pages made available offline are automatically updated in the background regardless of whether the user or Notion downloaded them.
  1. The answer repeated wording from this secondary comparison about automatic sync and a premium Notion plan.

    Zapier: Obsidian vs Notion

    What it does not establish: The attribution matches the source wording, but Notion documentation is needed to distinguish automatic offline downloads from background updates.

  2. Notion distinguishes manual offline downloads on all plans, automatic downloads on paid plans, and background updates.

    Notion Help: Use pages offline

    What it does not establish: These conditions concern Notion desktop and mobile apps. The documented offline-page feature is not available in web browsers.

A reusable evidence checklist

Use the following checklist on either product's answer. Record evidence first; make the recommendation afterward.

A compact evidence log can use these fields: claim | source URL | evidence level | supporting passage | limitation | checked date. It makes a review easier to inspect than a screenshot of a polished answer alone.

  1. Save the setup. Record the exact prompt, date, account state, plan, selected tools, and model or mode when visible.
  2. Separate access from quality. Mark a sign-in or retrieval failure as unavailable, rather than awarding the other product a quality win.
  3. Match each important claim to its URL. Record whether the source is directly relevant, background, or unrelated.
  4. Label the evidence level. Distinguish search snippets, page text, comment text, transcripts, and direct video inspection.
  5. Check decision-changing facts. Follow pricing, permissions, offline behavior, and other product restrictions to official documentation.
  6. Record missing coverage. Note inaccessible pages, unretrieved comments, absent transcripts, and limits on the available sample.
  7. Do not turn one task into a general ranking. Broader claims need multiple representative tasks and comparable access conditions.

What this comparison establishes

Kruqond produced a cited answer and disclosed important limits on its Reddit and YouTube evidence. The audit also found an indirect Reddit match and a Notion claim that needed a more precise primary-source qualification. Perplexity's answer quality remains unmeasured because the session did not produce an answer before sign-in.

If you are choosing a research assistant, try your own real task and inspect the evidence that changes your decision. For social research, the useful question is not just which platforms appear in the answer. It is what content was actually retrieved, how relevant it was, and whether the final claim stays within what that content supports.

Before you finish

  • Record the same prompt and access conditions for each product.
  • Distinguish source discovery from reading comments or a transcript.
  • Check decision-changing claims against primary documentation.

Related guides and product information

Follow the evidence, not just the answer.

Try five free searches without signing up, then open the cited sources yourself.

Try Kruqond