PaperPico

A paper you can submit. References from search, not memory.

Give PaperPico a topic and a template. It searches the literature, compares each reference with its registry record, writes the sections, works to your page count, and hands you a PDF with the LaTeX source. Usually 10 to 30 minutes, running in the background.

Write my first paper See what five trial runs produced See pricing

The sample papers on this page were written by PaperPico from the one-sentence briefs shown beside them. With no data from you it writes a literature review, a position paper or a method proposal and says so in the paper, unless you turn on the experiment add-on. Check claims, references and numbers before you submit.

Page 1 of a four-page IEEE conference paper: Event Cameras for Robot Perception, with abstract, index terms and introduction in two columns.

Sample output

Brief
Title
Template
Delivered

Reference search runs across these registries and indexes

Crossref
arXiv
DOAJ
INSPIRE-HEP
PubMed, via Europe PMC
OpenAlex

Research from universities around the world, in the indexes we search

Each logo links to an example indexed paper; it does not indicate an institutional endorsement.

What happens while you are away

You describe the paper once. It arrives in your project with every file it was built from.

Stop	What PaperPico does	What you find in the project
Plan	Reads your brief and settles the template, length, language, section list and a word budget per section.	The outline, and the settings it recognised from your sentence.
Search and verify	Searches Crossref, arXiv, DOAJ, INSPIRE-HEP, PubMed and OpenAlex. References are added from search results, never typed from memory, and each entry is checked against its registry record.	refs.bib, built from retrieved entries, and a count of how many matched a registry record.
Write	Writes the paper section by section, citing as it goes, in the template you asked for.	main.tex and one file per section.
Compile and fit	Compiles, counts the pages, and trims or extends to get close to the page count you asked for.	main.pdf, fitted to your target length.
Final checks	Looks for undefined citations, citations without a matching reference, and references nobody cites.	A checklist, plus a note of what you still need to supply.

A project, not a chat thread

A general-purpose chat assistant is good at drafting text. A paper you submit also needs sources that exist, a template and a length. That is the job PaperPico is built around.

PaperPico

A paper project, built from one brief

References come from searches across academic indexes and registries.
Reference metadata is checked against registry records, with matches reported.
Choose a template; PaperPico compiles and works towards your page count.
Receive a PDF, LaTeX source, bibliography and delivery checklist.

A general AI chatbot

A flexible conversation, guided by your prompts

Finding academic sources depends on the tool and how you prompt it.
References may need to be checked against their records separately.
Formatting and page count can take several rounds of prompting.
Assembling a PDF and source files may require additional tools.

The right column describes how a general-purpose chat assistant is typically used, not a test of any particular product or version; such tools differ and change.

References come from search, not memory

Every reference is added from a search result, never typed from memory. In a paper PaperPico writes, each entry is checked against its registry record before it goes into refs.bib, and the delivery checks report how many matched.

The same check is yours to run on any reference list, including one that did not come from PaperPico: entries it cannot confirm against Crossref or OpenAlex are flagged for you to look at.

What this checks: that a record exists and that title, authors and year match it. No record found does not prove a work is invented; it means you should look.

What it does not check: whether the cited paper supports the sentence that cites it. That judgement stays with you.

Reference checkIllustration

Ref. 1Record found. Title, authors, year: matchVerified
Ref. 2Record found. Title and authors match, year differsMismatch
Ref. 3Record found. Authors could not be comparedPartial
Ref. 4No matching record in Crossref or OpenAlexUnverified
Ref. 5Record found. Title, authors, year: matchVerified

What it will not do

The boundary is part of the product. These are written down because a paper you submit under your name should not depend on a guess about them.

Check experiment results against the run.
Without your data and without the experiment add-on, PaperPico writes a literature review, a position paper or a method proposal with no results of its own, and says which one it is. With the add-on, it is built to take results from the output of the code it runs and to state what was not obtained; the code and the run log come with the project, so check the numbers against them before you use them.
It will not sign the paper for you.
Samples carry "Anonymous Author(s)". Author names, affiliations and any data you hold are yours to add.
It does not score plagiarism or detect AI text.
Plagiarism and AI-detection checks are not part of PaperPico, and the page does not promise what it does not build.

Results of your own, when you want them

Turn on the experiment add-on (paid) when you start a paper, or later on a paper already delivered. PaperPico plans the experiment, runs it as a separate job in its own isolated sandbox for up to 90 minutes while the paper is being written, and fills the tables, figures and numbers into the paper when it finishes.

It covers classic machine learning on public datasets, simulation and numerical computation. The code, the run log and a report on how to reproduce the run come with the project, and the paper says the experiments were run by the system.

What it runs: experiments that need only public data or a model it can simulate, small enough for one CPU within the time limit.

If a run stops early: the writing step is instructed to state which results were not obtained, and the run log shows what actually ran. Check the numbers against the log before you rely on them.

Experiment add-onIllustration

PlanData, methods, protocol and outputs, in experiment_plan.mdWritten
RunIts own sandbox, fixed seeds, up to 90 minutesSeparate job
Resultsresults.json, tables and figures from the run's outputFilled in
RecordCode, run log and reproduction report in experiment/Included

Choose a format

Name one in your brief, or pick it before you start. The paper is typeset in that template from the first compile.

Publisher templates

IEEEConference and journal
ElsevierSingle-column journal layout
ACMTwo-column proceedings
LNCSSpringer proceedings

Generic layouts

Numbered referencesGeneric, single column
Author–year referencesGeneric, single column

Chinese

Chinese (ctexart)With GB/T 7714 references

English or Chinese. One to twelve pages. You set a minimum number of references and a page count. The page count is a target it works towards, not a guarantee.

Beyond the first draft

The paper is where it starts. The same project keeps going after delivery.

Check before you submit

Run the checks on the paper you are about to send, whether PaperPico wrote it or not.

Reference check against Crossref and OpenAlex records
Journal fit for the venue you have in mind
An AI review of the manuscript, as guidance to check against your own reading

Reference check

Recordfound

Titlematch

Authorsmatch

Yeardiffers

Bring a draft instead

Paste your own text or load a .txt, .md, .tex or .docx file. PaperPico reformats it for the template you choose and completes the citations. Check the changes and any new references before you submit.

For example

Polish my draft, keep its meaning, add missing references, and lay it out as a 6-page Elsevier paper in English.

Revise the delivered paper

Ask for a change on the same project: fewer pages, another template, one more section. Open any source file and download the PDF and the LaTeX source whenever you want them.

Source viewer: .tex with a jump to the PDF page, refs.bib as a table, build logs

Project files

main.pdfPDF

main.texsource

sections/*.texsource

refs.bibreferences

Editor, library and PDF chat

Work on a document in the editor with the tool rail beside it, keep papers in your reference library, and ask questions of a PDF. The Agents chat answers with cited sources and suggests follow-ups.

Also in the workspace

Agents chatcited answers

Document editortool rail

Reference librarysaved papers

PDF chatask a paper

What five trial runs looked like

Five new topics, one run each, written by an earlier build of deep mode and then checked independently of the agent that wrote them.

Internal evaluation, 5 November 2026, run on an earlier prototype build, not the current one: five topics, one run each, recompiled independently, no undefined citations, every DOI resolved and each remaining entry matched by arXiv title. A later round on five other samples resolved 117 of 117 DOIs. On the current build those five samples took 6 to 41 minutes each; the limit is 50 minutes per paper of up to 12 pages, proportionally more for 15 or 20 pages. Neither round is a ranking against other tools or a promise for your topic.
Topic	Template	Pages (target)	References	Found in a registry
Federated learning for medical image segmentation	IEEE conference	4 (4)	38	36 by DOI, 2 by arXiv title
Microplastics in freshwater ecosystems	Elsevier	6 (6)	44	44 by DOI
Large language models for legal text analysis (Chinese)	Chinese (ctexart)	4 (4)	36	35 by DOI, 1 by arXiv title
Energy cost of on-device LLM inference (position paper)	ACM	5 (5)	38	34 by DOI, 4 by arXiv title
Wildfire smoke and respiratory health	Single column, numbered	6 (6)	48	48 by DOI
All five		Target met 5 of 5	204	204 of 204

Describe the paper in one sentence.

Pick a template, a page count and a minimum number of references. You can leave while it works and come back to your project for the result.

Write my first paper