Frequently Asked Questions

Litra AI queries the OpenAlex academic database directly to retrieve papers. Unlike ChatGPT, which generates text, we fetch search results directly from real paper databases — so fabricated papers are never suggested. We also run link verification on every reference to confirm each paper exists.

Hybrid search combines keyword search with semantic search. Keyword search, like Google Scholar, matches exact terms, while semantic search finds papers with similar meaning or concepts. For example, a search for 'crop monitoring' would also discover papers about 'agricultural field observation' or 'farmland surveillance.' Litra AI merges both approaches so you never miss relevant papers.

Yes, absolutely. While Litra AI is an academic paper search service, it isn't limited to research use. Consultants use it to gather case studies as supporting evidence for client explanations, and businesspeople use it to back up proposals with citations. Because papers carry more credibility than web articles, they're valued as solid evidence in external-facing materials.

The papers suggested by Litra AI are retrieved directly from the OpenAlex academic database, so only real papers are displayed. AI summaries, analysis, and research assistant responses are generated based on paper abstracts. They're useful for quickly grasping the big picture — for deeper understanding, we recommend reading the original papers directly.

Litra AI accesses OpenAlex (280M+ papers). OpenAlex integrates over 50 data sources including arXiv and PubMed, covering 98.6% of PubMed and spanning a wide range of academic disciplines. Litra AI also calls the JaLC (Japan Link Center) API at search time, so papers from J-STAGE and other domestic Japanese journals are included in the search scope (toggleable from a checkbox in the search form, on by default).

Yes. Litra AI calls the JaLC (Japan Link Center) API at search time to find papers from J-STAGE and other domestic journals. It searches OpenAlex international papers in the same query, so domestic and international literature come back together. When you type a question in Japanese, the AI interprets its meaning and runs the domestic search alongside an English-language international search, then presents the results in Japanese. You can toggle whether domestic journals are included from a checkbox in the search form (on by default). Note that items without DOIs, dissertations, and conference proceedings fall outside this coverage — for those, we recommend also checking CiNii.

Results typically stream in within 1-3 minutes. Results appear progressively during the search, so there's minimal perceived wait time. Compared to Gemini Deep Research or ChatGPT Deep Research (5-30 minutes), Litra AI delivers results significantly faster.

Litra AI works with abstracts and publicly available information only — we do not access paywalled full text. To read full papers, use your institutional access or purchase individually. For open-access papers (arXiv, PubMed Central, etc.), we provide direct links to the full text.

The relevance score shows how well a paper matches your search, on a scale of 1 to 10. Papers scoring 8-10 are highly relevant and are specially highlighted in the search results. Papers scoring 5-7 are relevant, so check the content and judge for yourself. Papers scoring 1-4 have low relevance and are worth only a quick glance. The AI calculates the score from three factors: semantic match with your search, analysis of the abstract, and relevance of the title. The score does not indicate a paper's quality or importance, so a highly cited, well-known paper will still score low if it is far from your search topic.

Loosening your search conditions helps. First, try shortening your search text — packing in too many conditions narrows the candidates too far. Next, try a different way of saying the same thing. If a technical term isn't landing, rephrasing it in more common words also helps (for example, "LLM" to "large language model"). If you still find nothing, answering the AI's questions in more detail during the intent confirmation step will improve accuracy.

Making your search more specific improves the scores. Instead of "machine learning", describe your research topic in detail, such as "methods for improving accuracy in image classification using machine learning". Answering the AI's questions during the intent confirmation step improves accuracy further. You can also ask the research assistant on the results screen to "find papers with higher relevance".

You can sort by three criteria. The default is by relevance, which lists papers in order of the AI-calculated relevance score. Sorting by citations lists papers by how often they have been cited by other papers. Sorting by year lists the most recently published papers first. Use the buttons at the top of the search results screen to switch between them.

When your search could be read in more than one way, we confirm which meaning you intend. For example, searching for "bias" could mean statistical bias or cognitive bias, and the papers you find will differ. Just answer in your own words, and the AI will optimize the search based on your reply. If you are in a hurry, you can skip the confirmation and start the search right away.

The free tier gives you 3 credits (approximately 3 searches). GPT-5.4 mini is available. You can try the same features as paid plans — multilingual translation, AI analysis chat, and paper analysis. No credit card required.

Credits are the currency used for paper search and AI analysis on Litra AI. They are consumed based on the actual AI usage cost, so the amount varies with what you do. As a guide, a paper search alone costs about 0.8 credits. If you search and then exchange a few messages with the AI, one session costs roughly 1.6 credits with Auto or GPT-5.4 mini, 2.0 credits with Haiku 4.5, and 4.3 credits with Sonnet 4.6, depending on the analysis model you choose.

Yes, unused credits roll over to the next month. The maximum balance is capped at 2x your monthly credits. For example, on the Standard plan (100 credits/month), you can hold up to 200 credits including carried-over amounts.

We accept credit cards (Visa, Mastercard, American Express, JCB). All transactions are securely processed through Stripe.

Yes. We support reimbursement-based payment for KAKENHI. After paying by credit card, a receipt is automatically issued that you can submit to your university's accounting department. The receipt does not include a company seal (inkan), but this is legally fine. If your university requires a seal, please contact us and we'll accommodate. Bank transfer (invoiced payments) is planned for the future.

Yes, you can cancel at any time. After cancellation, you can continue using the service until the end of your current billing period.

If you purchased an annual plan within the last 14 days, you can request a refund from your account page. The refund amount is prorated based on unused credits (refund = payment × remaining credits ÷ total credits). Monthly plans do not offer prorated refunds, but you can continue using the service until the end of the billing period. A processing fee (3.6%) may be deducted from refunds.

We do not offer a dedicated enterprise plan. For research lab use, we recommend Pro ($69/mo, 400 credits) or Max ($199/mo, 1200 credits). If Max is insufficient, you can purchase multiple Max subscriptions.

If you have forgotten your password, you can reset it from "Forgot your password?" on the sign-in page. If the reset email does not arrive, check your spam folder and confirm you entered the same email address you registered with — delivery can take a few minutes. If you registered with a Google account, no password reset is needed; sign in with the "Sign in with Google" button. If Google sign-in returns an error, try clearing your browser cache and cookies, disabling your popup blocker, and then a different browser, in that order.