How AI research assistants handle paywalls, and legal ways to reach the full text

AI research assistants handle paywalls by reading what is open to everyone: titles, abstracts and open-access full texts. A paywalled paper's full text reaches the tool only through a publisher agreement, your institution's access or a PDF you upload yourself, and Elicit, Consensus and Semantic Scholar each describe their version of this on their own pages. When a tool has only the abstract, you can still reach the full text legally through an open-access copy, your library login, interlibrary loan or the author.

This guide explains those layers, what three well-known research tools say about paywalled journals, why a chatbot usually can't use your library login, and the legal routes to a full text. Everything about other products comes from their own help and product pages as we read them in October 2026, and we ran no tests on them. We make Barside, a free desktop app for Windows, Linux and Mac, and the later sections show how to keep the papers you can read in notes and ask questions across them with the AI you connect.

How AI research assistants handle paywalls: three ways in

AI research assistants handle paywalls in layers, and knowing which layer a tool reached tells you how far to trust its summary. Every tool starts with metadata: the title, the authors, the journal and usually the abstract. Google's Scholar help page says abstracts are freely available for most articles, while reading the whole article may need a subscription.

The second layer is open access. A paper published under an open license, or a manuscript the author posted in a repository, can be read in full by anyone, and that includes an AI tool. The third layer is the paywalled full text, and a tool reaches it in one of three ways:

• A publisher agreement. The company behind the tool licenses full text from publishers and reads it inside its own product.
• Your institution's access. The tool, or a browser extension that comes with it, uses the subscriptions your library already pays for.
• Your upload. You give the tool a PDF you already have, and it reads that file.

When none of these applies, the tool works from the abstract alone. An abstract can state a result more firmly than the data supports, and it usually leaves out the methods, the sample and the limits. So the first thing to check in any AI summary of a paper is which layer the tool read.

What Elicit, Consensus and Semantic Scholar say about paywalled journals

AI research assistants handle paywalled journal access in different ways, and Elicit, Consensus and Semantic Scholar each spell out their approach on their own pages. None of them claims to open every paywall. Each one points you to your own access for the papers it can't read.

Elicit

Elicit's copyright FAQ says its search uses metadata such as titles, abstracts and authors from both open-access and paywalled papers. Full text comes from open-access papers, papers you upload with appropriate access, or papers Elicit can read through its own publisher agreements. Its page on paper source labels tells you which case you are in: Full text (or Full text from Library when it matched a PDF already in your Library), Abstract only, or PDF link available. That last label means Elicit knows where the PDF is hosted but couldn't download it, usually because the publisher restricts automated access.

To fill the gaps, the labels page suggests three things: install the Elicit browser extension to fetch papers with your institutional access, upload PDFs to your Library, or download PDFs from publisher sites where you have access and upload them during Full Text Screening. The FAQ adds a caution that applies to any AI tool. Whether you may upload subscription articles depends on your organization's agreement with the publisher, and "Some require an AI or data mining license to upload content into an AI platform." Elicit recommends checking with your legal or library services team.

Consensus

Consensus's full-text feature page says Consensus reads the full text, including paywalled work from major publishers, and names partners such as Wiley, AAAS, Taylor and Francis, Sage, ACS and APA. For any open-access paper, it pulls the full PDF into Consensus. If a paper isn't in its index, you can upload PDFs or sync your Zotero library to bring papers in.

The feature page describes that reading as Consensus going through the methods, findings and discussion to judge whether a paper answers your question. We found no line on it saying you can open a paywalled PDF yourself. An older Consensus post, its guide to university access from July 2025, says "Consensus does not provide special access to paywalled journal articles." For paywalled papers, that post described a link to the publisher's page with metadata and the abstract, and a LibKey integration it planned from the 2025-26 academic year. Consensus's current LibGuide says a Full Text link on a paywalled paper takes you to the publisher's website to read it with your own credentials, that people at an institution that subscribes to LibKey can turn it on in their Consensus Settings, and that a checkmark in a citation means the full text was analyzed, while no checkmark means only the abstract was.

Semantic Scholar

Semantic Scholar's FAQ on reaching full text describes the access options under each paper's abstract: View PDF, View Paper or View via Publisher. If a paper isn't freely accessible, you may land on a publisher's website with options to buy it. When you see "No Paper Link Available", Semantic Scholar has no source link, and the page holds only the title and other metadata. The FAQ also says Semantic Scholar has partnered with Get Full Text Research (GetFTR) and LibKey to give direct access to subscription articles your institution has access to.

Why an AI chatbot usually can't use your library login

An AI chatbot's access to paywalled academic content runs into login limitations for a plain reason: your library login belongs to you, and the chatbot fetches pages on its own. A tool that opens a journal page without your sign-in sees what any signed-out visitor sees, which is often the abstract and an offer to buy the article. Unless a tool's own documentation says it uses your institutional access, as Elicit's browser extension does, expect it to read public pages only.

Barside's own tools work this way, and its guides say so. The research agent reads up to 5 public pages without your browser's cookies or sign-ins, so pages behind a login usually can't be read, and it skips PDFs. Saving a webpage into a note uses a separate session with none of your sign-ins, so a page behind a paywall usually can't be saved that way, and an address that downloads a PDF has no page to save.

Three habits keep you on safe ground with any chatbot or research tool:

1. Keep your library password out of chats. Sign in yourself, in your own browser. Barside's assistant, for one, is told never to operate password fields, and you type passwords yourself.
2. Check the license before you upload. A PDF you downloaded through your library can come with terms about AI tools. Elicit's FAQ notes that some publishers require an AI or data mining license, so ask your library before you upload subscription articles anywhere.
3. Ask what the tool read. If a summary of a paywalled paper has no full text behind it, treat it as a summary of the abstract and go get the paper before you quote it.

Library logins, LibKey and other login sources that open the full text

AI research assistants handling paywalled journals rely on login sources you already have: your library's subscriptions, reached on a campus network, through a library proxy, or through link services such as LibKey and GetFTR. The tools above connect to these in different ways, and knowing which one your tool uses tells you where to sign in.

• On campus. Google's Scholar help page says that on a campus network, access links automatically include your library subscriptions.
• Off campus. Scholar's off-campus access links use the subscriptions Scholar recorded while you were on campus, and those records expire after 30 days. Without them, you may need your library's proxy or a VPN.
• Link services. Semantic Scholar works with GetFTR and LibKey, and Consensus's LibGuide says people at an institution that subscribes to LibKey can turn it on by selecting their institution in Consensus Settings.
• Browser extensions. Elicit's FAQ says its extension goes to the publisher's website, finds the PDF and downloads it with your existing institutional access, and that the PDF passes through the extension without being stored locally.

If you have no institution, these routes are closed to you. That makes the open routes in the next section the ones to learn first.

Legal ways to reach the full text of a paywalled paper

When a tool shows only the abstract, try these routes in order. Each one is legal, and the first four usually cost nothing.

1. Look for an open-access version. Authors often post a manuscript in a university or subject repository. Unpaywall's browser extension listing describes it as a way to "Legally get full text of scholarly articles as you browse", pointing you to author-posted manuscripts on university and government web servers. The listing says papers posted outside publisher permissions are not in its index. In Google Scholar, look for a [PDF] link beside a result, or open All versions to see other sources.
2. Use your library. Sign in through your library's website or proxy, or use a computer on campus. For students and staff, the library's subscriptions are the usual way into paywalled journals.
3. Ask for interlibrary loan. If your library doesn't subscribe to a journal, ask a librarian whether they can request the article from another library.
4. Write to the author. The corresponding author's email address is often printed on the article's first page or the publisher's page. Ask for a copy, or for a version they are allowed to share.
5. Buy the article. Semantic Scholar's FAQ notes that a publisher's site may offer the paper for purchase. It's the last resort, and sometimes the quickest one.

How to keep the papers you can read in Barside

Once you have a paper legally, the work moves to reading it and finding it again. Barside is a free notes app for Windows, Linux and Mac where each paper can live in a note next to your own thinking, and its AI search answers questions from your notes with numbered sources. Barside isn't a reference manager, so keep your citations and bibliography in the tool you use for them now.

1. Download Barside and open it.
2. Make one note per paper, titled with the author and year, for example Okafor 2024. The literature review template adds a sources table to fill in as you read.
3. Drag the PDF into the note, or use the paperclip in the note toolbar. Barside shows its pages and keeps the file in the note, as a card you can open, with Show in folder and Save a copy.
4. Paste the passages you want to find later into the note as text, with the page number, and add two or three sentences in your own words.
5. For text you can't select, such as a scanned page, press Alt+Shift+S, drag across it and choose Copy text. The text is read on your computer; see copy text from the screen with OCR.
6. Link ideas across papers by typing [[ and picking a concept note, as the guide to linked notes and backlinks shows.
7. Open Settings → AI and connect Claude, ChatGPT, Gemini or OpenRouter. Under AI spending limit, turn on Limit what Barside's AI spends and pick an amount.
8. Open Library & briefing, type your question into Ask your entire library and press Ask library. Click a source's title under From your library to open the note it came from.

Step 4 carries the most weight. AI search works on text: a note's words, your Reading inbox, bookmarks with saved page text, call transcripts and screenshots. Barside's guides don't list the text inside an attached PDF among the things it searches, so the quotes you paste are what the AI can match and cite. For a paper published as a web page that you read while signed in, Library & briefing → Sources also has Save webpage content: choose the page's bookmark, paste the text and choose Save this text, and up to 120,000 characters are kept for search. Check your library's terms before you do.

Examples from research work

These are made-up examples that show how the routine fits real jobs.

• Example: a public health master's student. Lena's research assistant summarizes twelve studies, and its labels show that four were read from the abstract only. She finds open-access manuscripts for two of them, signs in through her library for the third and writes to the author of the fourth. Each paper gets a note with the PDF, her quotes and page numbers. When she asks her library "Which of my papers used a randomized design?", the answer cites her own notes, and she opens each one to check the quote.
• Example: a policy analyst with no library access. Marcus works at a small nonprofit with no journal subscriptions. He relies on open-access copies and emails to authors, keeps a note for every paper he could read in full, and marks the rest abstract only, so he never quotes a finding he hasn't seen in the paper.
• Example: a science reporter on deadline. Dana's chatbot cites a study for a striking figure. Its source turns out to be the abstract page, so she asks the university press office for the paper, files it in a note with the passage that holds the figure, and links the note to her draft.

Limits and privacy

Know the limits before you build a research routine on Barside:

• No way past a paywall. The research agent and saved webpages see what a signed-out visitor sees, and the research agent skips PDFs. Barside is for papers you can already read.
• Matching by words. AI search matches the words in your question, so "self-efficacy" won't find a note that only says "confidence". Use the terms your notes use.
• A small window. Each answer draws on the six best passages, with at most two from one note, so ask about one theme at a time.
• No reference manager. Keep citations and bibliographies in the tool you already use for them.
• Computers only. Barside runs on Windows, Linux and Mac, and there is no phone app.

Your library stays in Barside's data folder on your computer. When you ask a question, your question and up to six matching excerpts go to the AI company you connected, and that company's data policy applies, which is one more reason to check a paper's license before you quote it into notes you ask AI about. To keep a workspace or a note out of answers, untick it under Choose included workspaces & notes in Sources. Barside also counts which features you use: it sends the names of the features used that day, with your system and Barside version, never your notes, files or who you are, and a switch in Settings turns it off. Read how AI search uses your data and the AI you connect.

How it compares with research tools

Elicit, Consensus and Semantic Scholar are built to find papers and, where their access allows, read them for you, and that makes them good at discovery. Barside does another job. It holds the papers you can already read, the quotes you chose and your own notes, and it answers from them with citations that open your notes. Many researchers will want both kinds of tool: one for searching the literature, and one for what they keep.

To choose the rest of your setup, read the best AI note apps for PDFs and literature notes and the guide to AI research tools with citations and how to check them. AI search that cites its sources explains how Barside picks passages and checks every citation before you see an answer.

Works well with

These guides cover the pieces of the routine above, from asking your notes to reading text off a scanned page.