Best AI With Citations in November 2026: 12 Tools Scored on Sources You Can Check

Disclosure: NOXI publishes this page and is one of the twelve tools scored. That's why the rules, the weights and the source of every cell are printed, along with the readings that move NOXI up and down.

Three tools share the top score, and none of them leads it alone. Consensus, Elicit and Jenni AI are identical on all seven rules, so our printed tie chain runs level to the end and the alphabet sets their order. All three search a named database of research papers, show the exact passage behind a citation, export references for a reference manager in formats such as RIS or BibTeX, and document a guard against claims their sources do not support.

NOXI is seventh of twelve on 53, level with ChatGPT and listed first on the printed tie chain, where the rule on made-up sources reads 15 against 8. NOXI is a document maker that cites, built to hand over a finished page rather than to search papers. It takes full marks on three rules: sources you can click next to the claim (every fact keeps its source on the page, as an inline link, a footnote or a sources list), export that keeps the links (they stay clickable in PDF and Word), and a guard against made-up sources (when a fact cannot be confirmed, NOXI leaves a clear field for you to fill instead of guessing). Its free plan covers five documents a month, the middle mark on the free-plan rule. Its three zeros sit on the research-library rules, the exact passage, a paper database and citation styles: NOXI searches the live web and keeps each fact's source on the page.

Read the table another way and NOXI moves both ways. Drop the two academic rules and rescale, the general-web reading, and NOXI is fifth on 71, behind four tools on 91. Score NOXI's first rule at 10, because its record also allows a sources list, and it is tenth on 43. At the top, one cell is enough: read Elicit's no-answer sentence as a general warning and Elicit falls to 86, third. Read Perplexity's free plan by its plans article and Perplexity is fifth on 61, with NOXI still seventh.

Best AI with citations: the twelve at a glance

On seven documented rules checked on 5 November 2026, Consensus, Elicit and Jenni AI score 93 of 100, Gemini Notebook 68, and 53 points separate the top of the table from Google AI Mode at 40.

How the ties were broken. The printed chain is K1, then K7, then K4, then K2, then name. Consensus, Elicit and Jenni AI are level on every rule, so the alphabet decides, and their order says nothing about which is stronger. NOXI and ChatGPT both score 53 and both take 20 on K1; K7 decides, 15 against 8. NOXI's product record says it leaves a clear field when a fact cannot be confirmed, while OpenAI's help says search results and citations "can be incomplete, outdated, or incorrect". Named, not ranked: Microsoft Copilot for consumers scores 33 on the same rules; Microsoft's page says its Deep Research is being retired in the consumer Copilot app, starting 18 August 2026.

Who should pick what

Six needs, six answers, and five of them are other tools.

• If you need peer-reviewed papers with the exact sentence behind each claim then Elicit, because its home page promises "sentence-level citations from the underlying sources" and its help says it highlights the source inside the paper [2].
• If you want a yes or no across many studies then Consensus, because its meter shows whether research leans "Yes", "No", "Possibly", or "Mixed", and hovering a citation shows the exact quote [1].
• If you need to know whether later papers support or dispute a claim then Scite, because it classifies each citing paper as "supporting, contrasting, or mentioning" [5].
• If you are writing an essay with references in APA, Chicago or Harvard then Jenni AI, because its page lists "10,000+ citation styles" and export to .docx [3].
• If you want answers about today's web with a numbered link per claim then Perplexity, because each answer "includes numbered citations linking to the original sources"; ChatGPT search does the same on its free plan, which its pricing table shows with no limit label [6] [8].
• If you need a finished one-pager or report that keeps each fact's source clickable in PDF and Word then NOXI, because every fact keeps its source on the page and, when a fact cannot be confirmed, it leaves a clear field for you to fill [7].

Key numbers

How we scored: seven rules, 100 points, printed in full

Seven rules worth 100 points, applied in the same words to all twelve rows, with half marks fixed before scoring.

The half-mark rule: a middle level is the printed one: 10 of 20, 8 of 15 (half of 15, rounded up), 5 of 10. The tie-break chain: K1, then K7, then K4, then K2, then name. Every cell that is neither full nor zero says why, in the cell, in the full table below. One reading for every row: on K1, citations documented without saying where they sit take the list level, 10; on K5, a cited report exported to Word, PDF or Google Docs counts as an export with sources.

Why the academic rules carry 25 points. Google's first page for this search, captured on 5 November 2026, is academic from top to bottom, and the autosuggest tails are "for research", "for academic writing" and "essay writer". So K3 and K6 stay in. A reader who only wants cited answers about today's web should read the general-web reading below, which removes both.

Every cell: twelve tools on seven citation rules

Eight of the twelve take full marks for sources next to the claim, so K1 separates little; the order is set by the passage, the paper database and the citation styles.

Each total is the plain sum of its cells: Consensus 20 + 15 + 15 + 8 + 10 + 10 + 15 = 93; NOXI 20 + 0 + 0 + 8 + 10 + 0 + 15 = 53; ChatGPT 20 + 0 + 0 + 15 + 10 + 0 + 8 = 53. Scite's K5 and K6 are named blanks: its help centre would not open, and the pages that did name no export format and no citation style.

The gaps between rows you will compare

No gap in the middle of the table is a single rule.

• The top three minus Gemini Notebook = 25: K3 and K6. Gemini Notebook cites the sources you add to it, so it has no paper database, and it documents no citation style.
• Scite minus Perplexity = 1: K2 plus 15, K3 plus 7 and K7 plus 7 for Scite, against K1 minus 10, K4 minus 8 and K5 minus 10.
• Perplexity minus NOXI = 1: K3 plus 8 for Perplexity's Academic mode, against K7 minus 7.
• NOXI minus ChatGPT = 0: K7 plus 7 for NOXI, against K4 minus 7, because ChatGPT's pricing table shows Search on Free with no limit label.

Six readings that move the order

The general-web reading puts NOXI fifth, NOXI's K1 at 10 puts it tenth, and one cell at the top breaks the three-way tie.

• General-web reading (without K3 and K6, rescaled to 100): Consensus, Elicit, Gemini Notebook and Jenni AI 91, level on every remaining rule; NOXI 71 and ChatGPT 71, NOXI first on K7, so NOXI fifth; Gemini 61 and Perplexity 61; Claude 55; Google AI Mode 53 and Scite 53, AI Mode first on K4; SciSpace 27.
• NOXI's K1 at 10 (its product record also allows a sources list): NOXI 43, tenth, behind SciSpace on 45 and ahead of Claude on 41.
• Elicit's K7 at 8 (its limitations sentence opens with "try hard to", read as a general warning): Elicit 86, third; Consensus and Jenni AI 93. Consensus's K2 at 0 (the reading before its help centre opened): Consensus 78, third, and Elicit first.
• Perplexity's K4 at 15 (its plans article promises very high basic-search limits on Free, against the pricing page's "Limited"): Perplexity 61, fifth, Scite sixth. NOXI stays seventh.
• ChatGPT's K7 at 15 (if OpenAI documents what ChatGPT does when no source supports a claim): ChatGPT 60, fifth, and NOXI eighth. ChatGPT's K4 at 8 (reading "subject to your plan's usage limits" as a cap): ChatGPT 46, NOXI seventh. Claude's K1 at 20: Claude 51, ninth. Scite's K5 at 10 (65) and Google AI Mode's K7 at 8 (33) move no place.
• Equal weights (each cell as a fraction of its rule): Consensus, Elicit and Jenni AI 0.93, Gemini Notebook 0.65, Perplexity 0.51, then ChatGPT, NOXI, SciSpace and Scite 0.50, Gemini 0.44, Claude 0.37, Google AI Mode 0.36. The weights did not put NOXI seventh. Rubric K has no right-to-left rule, so no such reading applies.

AI that cites sources: four ways a citation can look

"Cites sources" covers four different promises: a link next to the claim, the exact passage, a database of papers behind the search, and a stated behaviour when nothing supports the claim. Only Consensus, Elicit and Jenni AI document all four.

A link next to the claim (K1). Eight of twelve document it. Perplexity: "Each answer includes numbered citations linking to the original sources". Jenni AI connects "every AI-suggested sentence to the precise location in your source PDF". Four document links without saying where they sit: Scite ("linked to its source"), SciSpace ("AI answers with citations"), Claude ("Every response includes citations") and Google AI Mode ("links to explore further").

The exact passage (K2). Six of twelve document it, and it is the rule that separates a citation you can check in moments from one you have to read a whole paper for. Consensus: "When you hover over an in-line citation, Consensus shows the exact quote". Elicit: "We then highlight the source of the content within the relevant papers." Gemini Notebook: "To get the full quoted text, hover over any citation." Scite: "See the exact text from each citing paper". Claude lists "Relevant quotes when appropriate". Perplexity's hover shows a label for the cited domain, which describes the site, not the passage.

A database of papers (K3) and a stated behaviour with no source (K7) have their own sections below.

Best AI for research with citations: the scholarly databases

Five of the twelve search a named database of research papers: Consensus, Elicit, Jenni AI, Scite and SciSpace.

• Consensus describes "a database of over 220 million peer-reviewed research papers", naming Semantic Scholar, OpenAlex and PubMed, in its help centre.
• Elicit searches "over 138 million academic papers from Semantic Scholar, PubMed, and OpenAlex", in a help article dated 30 September 2026.
• Jenni AI offers to "Search 200M+ papers without leaving your editor", and its search box names Semantic Scholar, PubMed, arXiv and CrossRef.
• Scite says its assistant is backed by "323M+ full-text articles, including paywalled content from 44+ publisher partners".
• SciSpace says its papers are "sourced from a 200Mn+ corpus".

Those counts are each vendor's own, measured in different units (papers, full-text articles, a corpus), so they are printed and not compared. Perplexity sits at 8: its Pro Search help lets you "narrow your search to specific content types (e.g., Academic, Finance, your own files)", a mode on web search, while its "Choose sources" help lists only Web, Org Files and None. ChatGPT, Gemini, Claude, Google AI Mode and NOXI search the web, and Gemini Notebook works from the sources you add.

If you want the database tools compared on study habits rather than citations, our ranking of AI tools for students scores free offers and student plans.

Does ChatGPT give sources? What its help pages say

Yes, when it searches the web: OpenAI's help says "Web search is available on ChatGPT Free", that you "Select a citation to open its source", and that you can "point to a citation to preview it." That is full marks on K1 and on the free plan.

Deep research goes further on export. OpenAI's article says you get "a structured report with citations or source links" and can download it "in multiple formats, including Markdown, Word, and PDF", which is K5. OpenAI's pages name no paper database, no passage display and no citation style, so K2, K3 and K6 are zero. On K7 the help page gives a plain warning: "Search results and citations can be incomplete, outdated, or incorrect." It does not say what ChatGPT does when no source supports a claim, so K7 is 8.

Why does ChatGPT give fake sources?

A reference written from memory, without a search behind it, can look right and point nowhere. The fix is in how you ask: turn on search, open each citation, and check that the page says what the sentence says. Our guide to why chatbots make things up and how to check walks through the checks, and our Perplexity vs ChatGPT comparison scores the two assistants on research beyond citations.

Gemini, Claude, Perplexity and Google AI Mode: how the big assistants cite

The four general assistants all cite web pages, and each documents a different weak spot on these rules.

• Gemini (46). Sources sit in line or behind a Sources button; a long direct quote adds "a link to that webpage in the sources list"; Export to Docs keeps the report. Its warning: "Gemini can hallucinate and present inaccurate information as factual."
• Claude (41). "Every response includes citations" and "Relevant quotes when appropriate", which earns K2. No page we opened places the citations on sentences or exports a cited answer.
• Perplexity (54). Numbered citations, an Academic search mode and research reports exported "to PDF or document". Its advice: "it's crucial to validate information by referencing the sources linked in your answer".
• Google AI Mode (40). Free in Search, with "links to explore further". It is one of seven rows with full marks on K7, because Google says AI Mode will "provide a set of web links if there's not high enough confidence".

For the two assistants side by side on many more rules, see our Gemini vs Perplexity comparison.

Best AI for finding sources for an essay or report

For an essay, finding a source is half the job; the other half is getting it into your reference list. Nine of the twelve document an export that keeps the sources, and four document citation styles: Consensus, Elicit, Jenni AI and SciSpace.

• Consensus exports paper details "to CSV or RIS file formats" for EndNote, Mendeley and Zotero, and formats citations in "APA, MLA, Chicago, Harvard, BibTeX, or AMA/Numeric".
• Elicit exports research reports "as a PDF or Word file, regardless of which Elicit plan" you use, and tables of sources "as RIS or BIB files", which its pricing page lists under Plus, a paid plan.
• Jenni AI exports "to .docx, LaTeX, or HTML", exports your library as .ris, .bib or .csv on Free, and lists "10,000+ citation styles including APA 7th, Chicago, Harvard".
• SciSpace offers "Export to CSV, XLSX, BIBTEX" and a citation generator.
• Gemini Notebook exports to Docs, and its saved notes keep "clickable inline citations". Gemini exports to Docs, Perplexity exports research reports "to PDF or document", and ChatGPT's deep research downloads as Markdown, Word or PDF.
• NOXI keeps each fact's source on the page, and links stay clickable in PDF and Word.

Scite, Claude and Google AI Mode document no export of a cited answer on the pages we opened. If what you are writing is a report rather than an essay, our ranking of AI document generators asks which tools put sources inside the file itself, and our ranking of AI business plan generators covers plans for lenders.

Best AI for fact checking: who says when it has no source

Seven of the twelve document what happens when no source supports a claim, four give only a general warning, and SciSpace says nothing on the pages we opened.

• Gemini Notebook answers from the sources you add: "If the answer isn't in the source material, it won't provide a response."
• Elicit says it tries to "err on the side of Elicit saying nothing rather than saying something wrong". The sentence opens with "try hard to", the closest call in the table, printed above as a reading.
• Jenni AI runs Reviews, which "analyzes every claim in your paper, cross-references your sources, and flags issues", with labels such as Unsupported and Contradicted.
• Consensus colours papers and citations "by stance", and its meter shows whether research leans "Yes", "No", "Possibly", or "Mixed".
• Scite classifies each citing paper as "supporting, contrasting, or mentioning".
• Google AI Mode falls back to "a set of web links if there's not high enough confidence".
• NOXI leaves a clear field for you to fill instead of guessing when a fact cannot be confirmed.

Perplexity, ChatGPT, Gemini and Claude give a general warning or tell you to check the sources, the middle mark.

What the free plans give you, and what the paid plans cost

Two of the twelve give cited answers free with no stated cap, ChatGPT search and Google AI Mode; eight cap the free plan; Scite and SciSpace have no free plan.

The middle mark, 8. Consensus Free gives "10 Pro messages per month" and lets you "Keep using basic paper search for free, without any AI analysis." Elicit's free Basic plan has "Limited usage for Research Agent and Research Reports". Jenni AI Free lists "5 AI Chat messages" and "3 Reviews". Gemini Notebook and Gemini run on "Standard limits" without a plan, and Gemini's limit refreshes every 5 hours. Claude's free limits "reset every five hours." Perplexity's pricing page marks web answers "Limited" on Free; its plans article lists Free at "Research Queries" "1/month" and "Pro Searches" "3/day". NOXI is free to start: the free plan covers five documents a month, and every change after that, by chat or by hand, is free. New accounts get a welcome gift to use in their first 14 days. Downloads carry no watermark, on the free plan too; no credit card is needed to start, and creating needs a free account.

Paid only, 0. Scite offers a "Start free 7-day trial", after which you are "automatically enrolled in the subscription plan you selected". SciSpace shows no free plan card; its pages offer a 100% money back guarantee for 24 hours.

Prices, as each page served them on 5 November 2026:

Elicit and Jenni AI were read with annual billing selected, and Perplexity's page prints only the annual Pro figure, so no monthly-billing price is given for the three. Gemini Notebook works without a plan on Standard limits, and Google AI Mode is free in Search.

What changed recently

Five vendor pages behind the cells above carry a date in the past three weeks, and the two from Perplexity describe what its free plan includes.

Perplexity's 26 October page does not make Deep Research free without limit: every plan gets some free usage, and credits are used once it runs out.

What these rules don't measure

The biggest thing seven rules leave out is whether a cited source actually says what the answer claims, and no vendor publishes an audit we could reuse.

• Citation accuracy. A documented link is not a correct link, and a stance label can be wrong. Every tool here deserves the same caution.
• How much of the literature each database covers. The paper counts above are each vendor's own claim, and none of them is a measure of recall for your field.
• Paywalled papers. Scite names full-text content from publisher partners, a real strength for anyone without a library login, and no rule here rewards it.
• Working from your own files. Gemini Notebook cites the sources you add, up to 50 for Free users, which is exactly what a student reading a set list wants.
• Writing quality, speed and privacy. None can be read off a help page, and all three matter.

Not measured for anyone: offline use, team plans, legal citation formats such as Bluebook, and accessibility. For slides rather than documents, our guide to AI presentations with research and citations asks the same question of decks.

What would change this ranking

Three dated, checkable changes would each move a named row.

• 1. Consensus, Elicit or Jenni AI changes one scored line. If Elicit's limitations article drops "err on the side of Elicit saying nothing" (K7 to 8), Elicit falls to 86, third. If Consensus's pricing page removes the 10 free Pro messages (K4 to 0), Consensus falls to 85, third. Pages: support.elicit.com/en/articles/14757928 and consensus.app/pricing. Re-check 1 December 2026.
• 2. Perplexity's pricing page drops "Limited" from Free on "Get answers from the web" (K4 to 15), or its "Choose sources" help names an Academic paper database (K3 to 15). Either one gives Perplexity 61, fifth. Pages: perplexity.ai/hub/pricing and help article 10354759. Re-check 1 December 2026.
• 3. OpenAI documents what ChatGPT does when no source supports a claim (K7 to 15): ChatGPT 60, fifth, and NOXI eighth. Or Anthropic documents citations attached to sentences (Claude's K1 to 20): Claude 51, ninth. Pages: help.openai.com article 9237897 and support.claude.com article 10684626. Re-check 1 December 2026.

Worked example: eight peer-reviewed sources by Friday

Example Daniel Mensah, a fictional second-year nursing student in Manchester, needs eight peer-reviewed sources on pressure ulcer prevention for an essay due Friday. His reference list must be in APA, his module leader wants each claim traceable to a sentence in a paper, and his budget is zero. The table turns those needs into four filters.

Filter one, papers rather than web pages (K3 = 15). Five rows pass: Consensus, Elicit, Jenni AI, Scite and SciSpace. Perplexity's Academic mode, at 8, is a search mode on the web, so it stays out.

Filter two, the exact sentence (K2 = 15). Four of the five pass: Consensus, Elicit, Jenni AI and Scite. SciSpace's page offers "Read paper abstract and verify claims", which is reading, not a highlighted passage.

Filter three, a free plan rather than a trial (K4 above 0). Three of the four pass. Scite starts with a 7-day trial that rolls into a paid plan, and his budget is zero, so it leaves.

Filter four, APA into his reference manager (K5 = 10 and K6 = 10). All three pass on the rule: Consensus exports RIS and formats APA, Elicit exports RIS or BIB, and Jenni AI exports .ris and .bib with APA 7th among its styles. At zero budget, note that Jenni AI's pricing page lists that export on Free, and Elicit's lists RIS and BIB export under its paid Plus plan.

So the three tools on 93 are the three that fit, and their free plans are small, so the arithmetic matters. Four questions in Consensus use 4 of its 10 free Pro messages this month and leave 6. If those four answers surface 14 candidate papers and he keeps the 8 whose quoted passage really supports his point, he discards 6, which is 43% of the shortlist. One Review of his finished draft in Jenni AI uses 1 of its 3 free Reviews and leaves 2 for the next essay. Elicit's free Basic plan states only "Limited usage", with no number, so he keeps it as the backup.

A general assistant would have given him links in seconds; he would still have had to find each paper and the sentence inside it. That is what K2 and K3 measure.

Build it in NOXI: a sourced one-pager with a clear field you fill by hand

NOXI does a different job from Daniel's tools: a finished page whose facts come from the live web with their sources, ready to hand over. This flow uses the One-pager format in NOXI 2.0, and its centre is step 4, the field NOXI leaves when a fact cannot be confirmed. NOXI 2.0 makes 14 formats from one request: presentation, document, brochure, social post, poster, one-pager, report, proposal, résumé, infographic, letter, invitation, certificate and menu, or Auto to let NOXI choose. The NOXI 2.0 launch article covers all fourteen.

Prompt 3 starts a second document, with the exported PDF attached as its input, and it is the second of five a month on the free plan. The clear field is the honest part of the flow: a blank you can see is safer than a number nobody can trace.

Frequently asked questions

What is the best AI with citations?

On seven printed rules checked on 5 November 2026, three tools share the top score of 93: Consensus, Elicit and Jenni AI, identical on every rule and listed alphabetically. All three search research papers, show the exact passage and export references. Gemini Notebook follows on 68, Scite 55 and Perplexity 54.

Which AI provides citations?

All twelve tools scored here document some form of citation. Eight attach sources to individual claims: Consensus, Elicit, Jenni AI, Gemini Notebook, Perplexity, NOXI, ChatGPT and Gemini. Scite, SciSpace, Claude and Google AI Mode document links without saying where they sit. Six also show the exact passage behind a citation. Checked 5 November 2026.

Is there an AI that gives real sources?

Yes. Consensus, Elicit, Jenni AI, Scite and SciSpace search named databases of research papers, so what they cite comes from an index of papers. Consensus, Elicit, Jenni AI and Scite also show the exact text behind a citation, which lets you confirm it quickly. Any tool can still misread a source, so open the link before you rely on it.

Does ChatGPT give sources?

Yes, when it searches the web. OpenAI's help says web search is available on ChatGPT Free and that you select a citation to open its source. Deep research reports carry citations or source links and download as Word or PDF. OpenAI warns that search results and citations can be incomplete, outdated or incorrect.

Why does ChatGPT give fake sources?

A reference written without a search behind it can look plausible and point nowhere. OpenAI's help page warns that even search citations can be incomplete, outdated or incorrect, and it does not describe what ChatGPT does when no source supports a claim. Turn search on, open every citation, and check the page says what the sentence says.

Does Perplexity cite its sources?

Yes. Perplexity's help centre says each answer includes numbered citations linking to the original sources, and research reports export to PDF or document. On 5 November 2026 its pricing page marked free web answers Limited, and its plans article listed one research query a month on Free. It names no paper database.

What is the best AI for research with citations?

For research papers, Consensus, Elicit and Jenni AI share 93 on our seven rules, checked 5 November 2026. Each searches a named paper database built on sources such as Semantic Scholar and PubMed, shows the exact passage and exports references for a reference manager. Scite adds whether later papers support or contrast a claim.

What is the best AI for finding sources?

For an essay, a tool that finds papers and exports them: Consensus exports RIS for EndNote, Mendeley and Zotero; Elicit exports RIS or BIB; Jenni AI exports .ris and .bib on its free plan. For current web facts, Perplexity and ChatGPT search attach a link to each claim. Checked 5 November 2026.

Which AI formats citations in APA?

Four of the twelve document citation styles on 5 November 2026. Consensus lists APA, MLA, Chicago, Harvard, BibTeX and AMA. Jenni AI lists more than 10,000 styles including APA 7th, Chicago and Harvard. Elicit exports RIS or BIB files for a reference manager, and SciSpace exports BibTeX and offers a citation generator.

What is the best AI for fact checking?

Look for a documented behaviour when no source supports a claim. Gemini Notebook says it gives no response when the answer is not in your sources; Jenni AI flags unsupported claims; Scite labels supporting and contrasting papers; Google AI Mode falls back to links when confidence is low; NOXI leaves a clear field to fill.

Sources & methodology

Every vendor page below was opened on 5 November 2026, signed out, with no cookie banner accepted, no form filled and nothing bought, between 15:35 and 15:54 UTC (19:35 to 19:54 in Abu Dhabi). Eighteen of them, the plan pages included, were opened again between 16:10 and 16:25 UTC; no scored line had moved. Pages that render by script were read in a real browser; Elicit's help centre and Claude's pages were read with a plain fetch. Quotes are fifteen words or fewer. Prices appear only in the currency the page showed. Daniel Mensah is fictional. NOXI is operated from Abu Dhabi.

• [1] Consensus: AI analysis, citations and visuals; Responsible AI limitations; Research database; Export to reference managers; Creating a citation; The Consensus Meter; Product changelog; Pricing, in US dollars; 15:50 UTC.
• [2] Elicit: elicit.com; Elicit's reliability; Elicit's source for papers, 30 September 2026; Export your data, 30 September 2026; Elicit's limitations, 16 September 2026; Changelog; 15:35 UTC; Pricing, in US dollars, 15:51 UTC.
• [3] Jenni AI: jenni.ai, including its "What's new" list; Pricing, in US dollars with Annual selected; 15:51 UTC.
• [4] Gemini Notebook: Gemini Notebook Help on citations, quoted text, sources, limits, export and answers from your sources; 15:47 UTC.
• [5] Scite: Features and Pricing, in US dollars, 15:51 UTC; Scite coverage at Research Solutions. Scite's own help centre did not open.
• [6] Perplexity: How does Perplexity work; What is Pro Search; Choose sources; Source labels; Research mode; Plans, 29 October 2026; Deep Research and Model Council in Computer, 26 October 2026; Pricing and Pro, in US dollars; 15:41 to 15:44 UTC.
• [7] NOXI: NOXI 2.0 product record of 4 November 2026, rows A (formats), B (research and sources), C (planning), D (styles), F (self-checks), G (editing), I (export), K (version history), M (Arabic), N (price), Q (no watermark) and R (no card); the NOXI 2.0 launch article of 3 November 2026; noxi.ai signed out, 5 November 2026.
• [8] ChatGPT: Searching the web with ChatGPT; Deep research in ChatGPT; ChatGPT pricing, served in AED; read in a real browser, 15:47 to 15:48 UTC.
• [9] Gemini: Gemini Apps Help on related sources, Deep Research, limits, exporting responses and accuracy; Google AI plans, US page in US dollars; 15:44 to 15:47 UTC.
• [10] SciSpace: Literature review and Pricing, in US dollars; 15:50 UTC.
• [11] Claude: Enable and use web search, modified 7 October 2026; Claude pricing, in US dollars; 15:39 to 15:40 UTC.
• [12] Google AI Mode: AI Mode in Google Search; 15:47 UTC.
• [13] Microsoft Copilot, named line: Transparency note, 18 August 2026; Deep Research in Microsoft Copilot; 15:47 UTC.