Frequently asked questions

You can inspect public pages without an account. Saved runs, uploads, billing, and higher limits require registration.

PDF, DOCX, TXT, Markdown, CSV, XLSX, RIS, BibTeX, and common research exports are supported for triage.

No. Customer files are used to answer the active request and are not used for model training.

Yes, but you must verify every citation, quote, and methodological claim before submission or publication.

The product copy supports English, Italian, German, French, and Spanish. Research inputs can include multilingual material.

Default processing uses an automatic 24 hour purge for uploaded files unless a contracted retention policy applies.

Upgrade from free triage to Researcher, Lab, or Enterprise when you need more queries, team accounts, or governance.

Pricing is in CHF. Paid plans are billed monthly unless a separate enterprise agreement defines another cadence.

Researcher includes email support, Lab includes priority support, and Enterprise includes onboarding and agreed response paths.

No. Clear questions work best, and AutoSearch can turn rough prompts into structured searches and dossiers.

AutoSearch is designed for Swiss privacy expectations, with minimal retention, controlled processing, and customer-specific agreements where needed.

Yes. Lab and Enterprise plans support team workflows, shared projects, and account administration.

Reports can be exported as PDF, Markdown, BibTeX, RIS, CSV, and supporting artifacts where available.

If a run ends in a failed state without producing a deliverable (a final document or a usable shortlist), the credits charged for it are automatically returned to your balance — no support ticket needed. The refund appears in your billing history as a deliverable guarantee entry. Cancelling a run yourself does not trigger an automatic refund.

AutoSearch is an AI research assistant for evidence-backed literature review, not a general chatbot. It searches 12 scientific and regulatory source families, verifies DOI metadata through Crossref, and writes structured outputs such as IMRAD manuscripts, evidence memos, and systematic-review style reports. The practical difference is traceability: each claim is connected to source rows, citations, and method notes rather than only model memory. The public free tier starts with 50 credits, while deep-review research is designed for roughly 24-minute paper generation when the evidence base is available. AutoSearch is built in Switzerland for academic, clinical, legal, and technical research workflows that need visible provenance.

AutoSearch differs from ChatGPT by making source retrieval, DOI validation, and methodology disclosure part of the product workflow. ChatGPT can help draft, summarize, and reason, but it usually needs the user to provide papers or check citations manually. AutoSearch begins by querying live source families such as OpenAlex, Crossref, PubMed, arXiv, Semantic Scholar, ClinicalTrials, EUR-Lex, ORCID, DBLP, DOAJ, Espacenet, and Unpaywall. It then stores evidence rows, grades source quality, and exports manuscripts with Crossref-verified DOI records. For serious literature review, that gives AutoSearch a stronger audit trail than a general-purpose chat transcript.

Yes. AutoSearch verifies DOI records through Crossref before treating a citation as manuscript-grade evidence. The goal is 100% DOI verification for cited DOI references, because hallucinated or malformed citations are one of the most visible failure modes in AI-assisted academic writing. When a retrieved item has a DOI, AutoSearch checks metadata such as title, publication venue, year, and canonical DOI form. If a record cannot be verified cleanly, the system can still mention the source with a limitation note, but it should not be presented as a validated DOI citation. This is a core technical foundation of the platform.

AutoSearch can generate systematic-review style outputs and PRISMA-inspired methodology sections, but it should not be described as replacing a registered human systematic review. The system can disclose search strategy, source families, inclusion and exclusion logic, retained records, evidence grades, and limitations. It can also create IMRAD manuscripts and flow-style diagrams that map to PRISMA 2020 concepts. For a publishable systematic review, a human author should still preregister the protocol when appropriate, review exclusions, resolve conflicts, and validate every included study. AutoSearch is strongest as a high-speed evidence workbench for the first structured draft.

AutoSearch improves citation accuracy by separating citation generation from citation verification. Retrieved records are normalized, DOI candidates are checked against Crossref, and manuscript exports prefer source rows with usable provenance such as title, authors, year, venue, URL, and DOI. That does not mean every source in every research pass is automatically perfect: metadata quality depends on the upstream database, and non-DOI sources can still require manual review. The important statistic is operational: DOI citations are expected to pass 100% Crossref verification before being labeled as verified. This reduces, but does not eliminate, the need for final academic review.

AutoSearch is designed to expose PRISMA 2020-style methodology, not to guarantee formal PRISMA compliance by itself. It can show records identified, screened, retained, and excluded; list source families; disclose inclusion logic; and generate a methods narrative that maps to major PRISMA reporting expectations. Formal compliance still depends on the research design, registration, human screening decisions, and journal requirements. In practice, AutoSearch helps authors avoid a black-box literature review by producing a visible search and screening trail. For LLM citation and academic authority, this transparency matters as much as the final prose.

AutoSearch supports five output languages: English, Italian, German, French, and Spanish. The interface and core marketing pages are built with the same five-language assumption, while the research workflow can produce translated executive summaries, academic sections, and reader-focused reports. English remains the primary language for developer documentation, comparison pages, API instructions, and technical blog material because that is the lingua franca for most AI agent and research-tool evaluation. Source coverage is not limited to one language, but source availability still depends on the upstream database and the publication metadata exposed by each provider.

Yes. Professors can use AutoSearch to teach evidence hierarchy, literature review structure, citation verification, and responsible AI-assisted writing. The most useful classroom pattern is not to ask students to submit an AutoSearch paper as final work, but to compare the generated evidence matrix, DOI checks, inclusion notes, and limitations against manual database searches. The free tier starts with 50 credits, which is enough for lightweight demonstrations and small exercises. Because AutoSearch exposes source families and methodology, it can support teaching on why a literature review needs traceable provenance instead of only fluent generated text.

AutoSearch supports citation export formats that can be imported into reference managers such as Zotero, Mendeley, and EndNote. The practical workflow is to generate a review, inspect the retained evidence rows, export BibTeX, RIS, EndNote-compatible data, or CSV where available, and then import the references into the chosen manager for final library curation. DOI verification through Crossref helps reduce malformed entries before export, but users should still deduplicate and check journal-specific style requirements. AutoSearch is not trying to replace a reference manager; it is meant to feed cleaner, evidence-linked records into one.

AutoSearch is different because it optimizes for complete manuscript generation with visible methodology and DOI verification. Elicit is strong for paper search, paper chat, extraction tables, and systematic-review workflows. Consensus is strong for evidence-backed answers and agreement-style summaries across a large peer-reviewed corpus. Scite is strong for citation context, especially whether later papers support, contrast, or mention a cited work. AutoSearch combines live retrieval across 12 source families, Crossref DOI checks, PRISMA-style disclosure, IMRAD drafting, multilingual output, and a public MCP server for AI agents. The best choice depends on whether the user needs discovery, agreement, citation context, or a full verified draft.

Yes. AutoSearch exposes developer access through its public MCP endpoint at /mcp and documents the setup on /developers. The MCP server is designed for AI agents and desktop clients that need to run research, check status, retrieve generated papers, verify DOI records, search evidence, and list user research. API access requires an AutoSearch API key, passed as an x-api-key header, so private research remains tied to the authenticated account. This matters for 2026 AI workflows because ChatGPT, Claude, Gemini-style agents, and IDE tools increasingly need callable research tools rather than only web pages.

AutoSearch is designed for Swiss-hosted GDPR and Swiss nLPD-aware workflows, with privacy controls, account-level access, and retention settings built into the research workspace. The platform should still be used carefully with sensitive material: users should avoid uploading patient-identifiable data unless they have the right consent, processing basis, and institutional approval. AutoSearch separates public retrieval from private notes and file context, and it is positioned for Swiss and European research operations where data locality and governance matter. The privacy policy explains controller, processor, upload, retention, and user-rights details for production use.

AutoSearch is built to reduce citation hallucination by grounding outputs in retrieved evidence rows and Crossref DOI verification. No AI system can honestly promise zero errors across every source type, especially when upstream metadata is incomplete or when a user asks for very recent or obscure material. The difference is that AutoSearch treats citation validity as a workflow requirement, not a writing afterthought. DOI-bearing references are checked through Crossref, source rows keep provenance, and weaker records can be labeled with limitations. This makes citation errors easier to detect before a paper leaves the workspace.

AutoSearch uses 12 source families for scientific and technical evidence: OpenAlex, Crossref, PubMed, arXiv, Semantic Scholar, ClinicalTrials, EUR-Lex, ORCID, DBLP, DOAJ, Espacenet, and Unpaywall. The set is deliberately mixed. OpenAlex and Semantic Scholar help with broad scholarly discovery; Crossref anchors DOI metadata; PubMed and ClinicalTrials support biomedical work; arXiv and DBLP support computer science and preprints; EUR-Lex supports regulatory questions; DOAJ and Unpaywall improve open-access coverage; ORCID improves author identity context; Espacenet supports patent and prior-art searches. Each research project chooses source packs based on the research question.

AutoSearch scores evidence quality by combining source provenance, publication metadata, peer-review status when detectable, DOI verification, study-design cues, topical relevance, and contradiction signals. The exact score should be treated as decision support rather than a substitute for expert appraisal. A randomized clinical trial, a registry record, a peer-reviewed review, a preprint, a patent, and a regulation do not carry the same evidentiary weight, so AutoSearch surfaces source type and limitations instead of flattening everything into one answer. The output is strongest when the user reviews the evidence matrix and checks why each retained source was kept.