AI Detector for Text, Images & Citations | ProofVerity
ProofVerity
ES Sign in Analyze a document

AI content detector for text, images and citations Proof, or it didn't happen.Human-written · 99%

ProofVerity is an AI detector for documents and images. It reads a file the way a forensic analyst would and marks AI-written passages, altered or generated images and citations that don't hold up, on the page, with the evidence attached.

PDF · DOCX · JPG · PNG · URL — exports as annotated PDF
manuscript_v3_final.pdf · 18 pages
AI text · 2 Citations failing · 2 Images altered · 1
Report ready · 00:41
4 / 18
Preprint · not peer reviewedp. 4 / 18

Nighttime Cooling Deficits in Mid-Sized European Cities, 2015–2024

L. Marchetti, D. Okafor, S. Lindqvist — Department of Urban Climatology

Abstract

We analyse nighttime surface temperature records from 42 mid-sized European cities between 2015 and 2024. Cities with less than 12% tree canopy cooled on average 1.8 °C less overnight than comparable cities above 25% canopy. It is important to note that these findings underscore the critical importance of urban greening initiatives, which play a pivotal role in fostering resilient and sustainable communities in an ever-changing climate.

1. Introduction

Nighttime temperatures matter more for health outcomes than daytime peaks [2]; the body recovers during cooler hours, and sustained warm nights are associated with excess mortality among older residents [3]. Previous work has concentrated on capital cities; mid-sized cities remain understudied despite hosting 38% of the EU population [7]. In this paper, we delve into this gap by leveraging a comprehensive multi-source dataset and a robust analytical framework.

thermal camera still — figure 2
Figure 2. Thermal camera still, Padua, 02:14, 21 July 2023. Surface temperature scale in °C.

2. Data

Canopy cover was derived from the Copernicus Land Monitoring Service High Resolution Layer [1], and nighttime temperatures from national meteorological archives, gap-filled with satellite land surface temperature where station density fell below one per 40 km².

Sample report: six findings on one page. Two AI-written spans (94% and 81%), one fabricated reference [7], one misquoted reference [3], one verified reference [1] and one altered image (2.1% of the frame). The manuscript is fictional; the failure modes are not.
Things worth checking
the rental listing photo a screenshot of a tweet the five-star review your kid's essay a still from the apology video the quote in the press release the photographer's portfolio that PDF from a stranger

Not only for professionals. Paste a link, drop a screenshot or share from your phone, and read the result in plain language, with the evidence one tap away. The first document needs no account.

§01 — Detectors

Four detectors.
One annotated page.

Each finding is pinned to the exact span, pixel region or reference it concerns. Nothing is folded into a single score you would have to take on faith.

AI-written01

Text written by a model

Sentence-level detection tuned for academic and journalistic prose. Reports a calibrated probability together with its error rate. Never a verdict.

Citation02

References that don't hold up

Every citation resolved against thousands of journals, books, archives and registries. Missing, misquoted, misattributed or invented: each is named as such.

Image · synthetic03

Pixels a model made

Generator fingerprints and diffusion artefacts read from the image itself. Content credentials and metadata read alongside, not instead.

Image · altered04

Retouched, spliced, inpainted

Clone stamping, splicing and re-compression traces, localised to the region they affect. You see where, not just whether.

§02 — Why the egg

In the early days of image generation, a convincing fried egg was a speedrun. People timed how long a model took to get one right. Yolk in the wrong place. Whites like plastic. A fork with nine tines.

Then one morning the eggs were perfect, and nobody could tell.

That is the morning ProofVerity is built for. What was obviously fake last year is invisible this year. Somebody has to keep looking.

fig. 0 — a fried egg. Drawn by a person, in about four minutes.
§03 — Workflow

Drop it in.
Read the margins.

A full report on a forty-page document takes under a minute. Reading it takes as long as you decide it should.

  1. 01

    Drop it in

    PDF, DOCX, images or a link. Share from your phone, drop a folder from the desk, or hand off an assignment from the LMS.

  2. 02

    Four detectors, one pass

    Text, citations and images are analysed in parallel. Every finding is pinned to its exact position on the page, with the evidence that produced it.

  3. 03

    Review, then export

    Accept or dismiss findings, add notes, export an annotated PDF with an evidence appendix, or push results to your CMS or LMS.

§04 — Who it's for

Built for people whose job is not to take your word for it.

Newsrooms and universities have whole departments for this. Everyone else has a phone, a forwarded screenshot and a bad feeling. Same detectors, same evidence.

Newsrooms

Photos from a tip. Quotes in a press release. A freelancer's sources.

Checked before the editor asks, with a chain-of-custody log you can hand to legal.

  • →CMS plugin and desk watch folders
  • →Alerts in Slack or Teams
  • →Custody log on every asset
Universities

A thesis read with every citation resolved.

AI prose flagged sentence by sentence, with evidence a student can see and contest. Thresholds set per department, not per panic.

  • →LMS integrations and bulk assignment checks
  • →Department-level thresholds and policies
  • →Student-facing evidence view
Everyone else

The listing photo. The screenshot your uncle forwarded. The essay your kid swears they wrote.

Paste a link or share from your phone. Results in plain language, with the evidence one tap away. No jargon, no dashboard.

  • →Browser extension and mobile share
  • →Plain-language results, evidence on tap
  • →Three free credits a month
§05 — Method

Evidence, not a score.

Detectors are wrong sometimes. Ours say how often, show their work, and leave the decision with the person whose name is on it.

Calibrated

Probabilities are calibrated per document type and shipped with their false-positive rate. A 94 means 94.

Contestable

Every flag carries the evidence that produced it, in plain language. People decide. The model argues its case.

Private

Processed in-region, deleted after 30 days, never used for training. Your documents are evidence, not data.

§06 — Pricing

One credit.
One document, or four images.

Every credit runs all four detectors; images cost more than pages because pixels take GPUs. One retracted story or one wrongly accused student costs more than a year of checks.

1 creditone document, up to 100 pages
1 creditfour images, any resolution
250 creditsa month in Professional: eight documents a day
Free1 seat
€0forever
3 credits a month
no overage

Three documents or twelve images. Enough to know before you reply.

  • →All four detectors
  • →Plain-language results
  • →7-day history, no export
Start free
Most chosen
Professional1 seat · +€12 each
€49per month · €45 billed yearly
250 credits a month
then €0.25 a credit; €0.20 past 5,000, €0.17 past 25,000

A class set of essays every week and the images that come with them, on one plan. Evidence that survives a follow-up question.

  • →Annotated PDF export
  • →Full source list for every citation
  • →API access
  • →Browser extension and mobile share
Start Professional
Desk & Department25 seats · +€9 each
€499per month · €459 billed yearly
2,500 credits a month, pooled
then €0.20 a credit; €0.17 past 25,000

One desk or one department, one bill, one set of thresholds.

  • →Everything in Professional
  • →CMS and LMS integrations
  • →25 seats, shared credits
  • →Priority support
Talk to sales
Campus & Newsroomunlimited seats
Customyearly contract
from €2 per student per year
from €1,200 per newsroom per month

A whole university or a whole title, on one contract.

  • →Everything in Desk
  • →Unlimited seats
  • →Dedicated hosting on request
  • →Invoicing and a yearly contract
Request a quote
Pay as you go

No subscription, API included. The price drops as you go.

Billed at the end of the month for what you used, from the web or the API. Each tier applies only to the credits inside it, so crossing a line never costs you more.

Credits 1–10 €0.99 each. The first one needs no account.
Credits 11–100 €0.90 each
Credits 101–1,000 €0.60 each
Credits 1,001–5,000 €0.40 each
Credits 5,001–25,000 €0.20 each
Credits 25,001+ €0.17 each. The floor.
Add-ons Certified report, signed and timestamped · €19 Analyst second opinion within 24 h · €99
What it would cost you

{{ calcDocs }} documents and {{ calcImgs }} images a month, {{ calcSeatsLabel }}.

= {{ calcCredits }} credits a month
Annual billing: one month free.

Example: 40 documents and 20 images a month are 45 credits. Pay as you go, €41.40; on Professional, €49 with 205 credits to spare.

A credit is one document of up to 100 pages, or four images. On any plan, credits past 5,000 in a month cost €0.20 and past 25,000 cost €0.17. Prices follow the billing toggle above.

{{ r.name }}Best fit
{{ r.cost }} {{ r.note }} {{ r.perPage }}
§07 — At a glance

ProofVerity, in one screen.

The facts an editor, a procurement office or a search engine needs, without the adjectives.

What it is
An AI detector for text, images and citations, with the evidence for every finding shown on the page.
What it detects
AI-written text at sentence level; AI-generated images; retouched or altered images; citations that are missing, misquoted, misattributed or invented.
Formats
PDF, DOCX, JPG, PNG and public URLs, from the web app, the browser extension, the mobile share sheet or the API.
Languages
Spanish, English and the other major European languages. The site is available in English and Spanish.
Output
Annotated PDF with an evidence appendix, JSON through the API, and CMS or LMS integrations.
Pricing
Free: 3 credits a month. Professional: €49 a month for 250 credits. Desk & Department: €499 a month for 2,500 pooled credits. Pay as you go from €0.99 down to €0.17 a credit. A credit is one document of up to 100 pages, or four images.
Accuracy
Calibrated probabilities with a published false-positive rate per document type. No single score; every flag carries its evidence and can be contested.
Data
Processed in-region (EU or UK), deleted after 30 days, never used for training.
Who makes it
FUNNELD LTD, 82A James Carter Road, Mildenhall, IP28 7DE, United Kingdom, Companies House no. 14582937. Contact: contact@proofverity.com.
Where
proofverity.com in English, proofverity.com/es in Spanish; the app runs at analyze.proofverity.com.
§08 — Questions

Questions people ask.

What counts as a credit?
One document of up to 100 pages, or four images. The credit is spent when the report is ready, not when you upload.
Is a 94% a verdict?
No. It is a calibrated probability with a published error rate, and every flag shows the evidence behind it. People decide.
Which formats can I check?
PDF, DOCX, JPG, PNG and public URLs. Batches and the API accept the same formats.
What happens to my documents?
Processed in-region, deleted after 30 days, never used for training. Delete them earlier from your history whenever you like.
Does it work in Spanish?
Yes. Text detection and citation verification cover Spanish, English and the other major European languages.
Is ProofVerity an AI detector or a plagiarism checker?
An AI detector and a citation verifier. It does not compare essays against other students’ work; it tells you whether text reads as machine-written, whether an image was generated or altered, and whether each citation exists and says what it is claimed to say.
Can I cancel?
Any time. Monthly plans stop at the end of the period. No lock-in, no cancellation fee.

Send us the one
you're not sure about.

Drop a PDF, DOCX or image here
Three free credits a month, no card. Then €0.99 a check, less as you go, or €49 a month.

One cookie keeps you signed in. With your permission, anonymous analytics tell us which pages help. No advertising trackers, ever. Cookie policy