To check one document, such as an essay or a draft, use a hosted checker such as Pangram, GPTZero, or Originality.ai. To mark likely AI-written posts, email, and articles as you read, without sending any text off your device, use an on-device filter such as Botfilter.
The hosted checkers are web services. You paste or upload text, or their browser extension sends it, and a score comes back. Botfilter scores the pages you already read with a model that runs in your browser, and it needs no account.
This comparison covers Pangram, GPTZero, and Originality.ai. Every vendor fact below comes from that company's own pages as checked on September 28, 2026.
Which AI detector fits your job
- One document you were handed, such as a student essay or a freelance article: Pangram, GPTZero, or Originality.ai.
- A quick check without signing up: GPTZero or Originality.ai.
- AI and plagiarism checks together, in volume: Originality.ai or a paid Pangram plan. GPTZero sells a plagiarism check as well. Botfilter does not check for plagiarism.
- Feeds, inboxes, and articles as you read, without text leaving your device: Botfilter.
Then ask what a wrong result costs you. A flag on a student essay has consequences for a person, so a score from any of these tools is one piece of evidence, not a finding. A folded post in your feed costs you one click to show it again.
AI detector comparison at a glance
| Tool | Where text is scored | Account | Free use | Lowest paid plan | How you give it text |
|---|---|---|---|---|---|
| Pangram | Hosted service; its pages describe no on-device scoring | Needed for the free plan; no payment method | 2,000 words a day, 3 image scans a day | Individual, listed at $20 a month for 300,000 words a month; see its pricing page for monthly and annual billing | Paste, upload PDF, DOCX, or RTF, browser extension, Google Docs, learning platforms, API. Gmail and automatic feed scanning are on paid plans. |
| GPTZero | Hosted service; its pages describe no on-device scoring | Not needed under 10,000 characters; a free account raises the limit to 150,000 | Scans under 10,000 characters | Professional, billed monthly or annually; its pricing page showed no price on the check date | Paste, upload, Chrome extension, Google Docs, Canvas, Google Classroom, Zapier, API |
| Originality.ai | Hosted service; its pages describe no on-device scoring | Not needed for 3 scans a day | 3 scans a day, up to 2,000 words each, 100-word minimum | Pro, $14.95 a month, or $12.95 a month billed annually, for 2,000 credits (1 credit is 100 words) | Paste or upload, Chrome and Firefox extensions, Google Docs, Moodle, Google Classroom, API |
| Botfilter | Your browser, on your device | None, and no sign-in | Free, with no daily quota | None | Reads the pages you open; there is no box to paste text into |
Sources: each company's pricing, product, and FAQ pages, listed under this article. Pangram's home page describes its free plan as “twenty free checks a day”; its pricing page states 2,000 words a day, and the table uses the pricing page. Prices change; check the vendor's page before you buy.
Pangram vs GPTZero vs Originality.ai
Pangram vs GPTZero
GPTZero is the quicker one-off check: paste under 10,000 characters and scan, with no account. Pangram asks for a free account but offers more ways in, including uploads of up to 100 files at a time, learning platforms such as Canvas, Moodle, and Brightspace, and an image detector. Pangram says user submissions are not in its training data. GPTZero’s pricing and FAQ pages do not state whether it trains on them.
GPTZero vs Originality.ai
Both let you check text without an account. Originality.ai publishes its price per word: $14.95 a month buys 2,000 credits, enough for 200,000 words of AI checks. GPTZero's pricing page showed plan names but no prices on the check date. Originality.ai may use your scans for training unless you opt out, while GPTZero says it uses stored dashboard data in aggregate to improve its service.
Pangram vs Originality.ai
For volume, Pangram's Individual plan lists 300,000 words a month at $20 a month. Originality.ai's Pro plan covers 200,000 words of AI checks for $14.95 a month, or 100,000 words when each check includes plagiarism, which costs 2 credits per 100 words. Pangram needs an account even for free checks, and Originality.ai gives 3 scans a day without one. For more on Pangram, our Pangram comparison sets it against Botfilter.
Where your text goes with each detector
Every hosted detector has to receive your text to score it. That matters when the text is private: an unpublished manuscript, a client email, a student's work under a school's data rules. Each company publishes what it does with that text, and the answers differ.
| Tool | Stored | Used for training | Security claim |
|---|---|---|---|
| Pangram | Text from the dashboard and extension is stored to show your history while your account is active, and you can delete it there. The extension also stores feed data by default for its Feed Health feature, with an opt-out. | Pangram says its detector is built on a proprietary dataset that does not include user submissions. | States SOC 2 Type II compliance. |
| GPTZero | Dashboard inputs are stored. The FAQ says documents sent through its API are not stored. | Not stated on its pricing or FAQ pages. The FAQ says stored data is used in aggregate to improve the service. | States SOC 2 Type II compliance, and that its systems are designed to comply with GDPR. |
| Originality.ai | Scan history is kept for 30 days on Pro and 365 days on Enterprise, and reports can be shared. You can delete scan history and request deletion of personal information. | Scans, including paid ones, may be used for training unless you opt out. The free checker page says opting out needs an account. | No certification named on its pricing, privacy, or FAQ pages. |
| Botfilter | No page text is stored or sent. It keeps settings, hashes of scored text, scores, and email thread IDs in your browser. | No. It never receives your text. | Scoring makes no network requests, confirmable in developer tools once installed. |
Both GPTZero and Pangram also mark text on the pages you read. GPTZero advertises its Chrome extension for automatic detection on Gmail, Google Docs, Classroom, and social media. Pangram's free plan includes an extension for any website, and its paid Individual plan adds automatic feed scanning on sites such as X, LinkedIn, Substack, and Medium, plus a Gmail integration. Both describe these as features of their hosted services, and neither describes scoring on the device. With Botfilter, the text, the model, and the score stay on your device, with no account and no cost.
The walkthrough of how Botfilter runs on your device covers what it stores, including synced settings, and how to check its network use once it is installed.
What each company reports about accuracy
Each company reports accuracy on its own test data, with its own mix of writing styles, AI models, and text lengths. We have not tested the hosted tools on the same texts as Botfilter, so this page does not rank them. The figures below are each company's own claim, and they cannot be compared with each other.
| Tool | Reported result | Measured on | Published in |
|---|---|---|---|
| Pangram | AUROC of 0.9916, false positive rate of 0.0041%, false negative rate of 0.3396% | Pangram's own evaluation | Pangram 4 technical report, July 2026 |
| GPTZero | Caught 95.7% of AI texts and flagged 1% of human texts as AI-written | The RAID benchmark | GPTZero home page |
| Originality.ai | 97.8% overall accuracy for its multilingual model, across 30 languages | Originality.ai's multilingual accuracy study | Originality.ai free AI checker page, which also says to treat a score as one signal, not a final decision |
| Botfilter | At Balanced, flagged 0.5% of human news articles, 1.6% of human work emails, and 1.8% of human web pages; caught 95% of text from the Claude family and 96% from the GPT family | A held-out test set: texts from the same sources it trained on, never seen in training, scored once | How we trained Botfilter |
Our explainer on AI detector accuracy covers what false positive rates mean for a real reader, and what published research found when detectors were tested independently.
Botfilter's model was trained and tested on English only. It scores other languages too, but those scores are untested and can be wrong. It skips text under 25 words. Its false positive rate on human X and LinkedIn posts is unmeasured, because no licensed set of those posts was available for testing. Human writing can be flagged, and text that a person and a model both edited is often missed. A flag is a probability, not proof of who wrote something.
What an on-device filter looks like on a page
Botfilter is for readers. It has dedicated layouts for X, Reddit, LinkedIn, Gmail, YouTube, and Hacker News, and it checks paragraphs on every other site. You choose what happens to text it flags as likely AI-written: Outline marks it in place, Collapse folds it to a one-line label you can click to show, and Hide removes it. You also choose how much it flags: Fewer, Balanced, or More. By default, hovering a flag shows its probability and a Not AI button, and Botfilter remembers that correction on your device. It never reads text you are typing in a compose box or form.
I got rejected from 47 jobs in 2019.
Today I run a team of 60.
Here's what nobody tells you about rejection:
It isn't a stop sign. It's a redirection.
Every no taught me something: → Resilience is a muscle, not a trait. → Your network is your net worth. → Growth lives outside your comfort zone.
The truth is, the most successful people I know didn't avoid failure. They embraced it, learned from it, and kept going.
So if you're facing rejection right now, remember this: your breakthrough might be one application away.
Agree? ♻️ Repost to help someone in your network who needs to hear this today.
We're hiring a second-shift baker. Four nights a week, 10pm to 6am, and yes, you get to take home whatever doesn't sell. You don't need a résumé. Come by Thursday afternoon and bake something with us. Tell anyone who might like the quiet hours.
Excited to share that our team just wrapped up an incredible quarter. 🎉
This journey has been a true testament to the power of collaboration, innovation, and a shared commitment to delivering value for our customers.
A few key takeaways: • Alignment across teams is everything • Customer feedback is a gift • Small wins compound over time
I'm incredibly grateful to work alongside such talented, passionate people who show up every day ready to make an impact. Here's to continuing to push boundaries and redefine what's possible in the months ahead.
Last week our warehouse ran its first full inventory count without overtime. It took three tries to get the process right. The first attempt fell apart because nobody owned the spreadsheet. The second time we split the building into zones and gave each shift lead one zone and one scanner. By the third count people were finishing early and asking what else needed doing. Nothing fancy, only clear ownership and fewer handoffs. Sharing in case another ops team is stuck in the same loop.
For eleven years, one person has kept the Gull Point crossing on schedule with secondhand parts and a notebook.
Ruth Anand gets to the dock at 4:40 every morning, forty minutes before the first crossing, and starts the ferry's engine by hand when the cold makes the starter sulk. She has done this for eleven years. In that time the boat has missed nine scheduled runs, and she can tell you the reason for each one.
Ferries play a vital role in connecting island communities to the mainland. Beyond their practical function, they serve as a lifeline that fosters economic growth, supports local businesses, and strengthens the social fabric of the region. In many ways, the ferry is more than a mode of transportation; it is a symbol of resilience and community spirit.
The boat itself is older than Anand. It was built in 1971 for a river crossing two states away and bought secondhand by the county in 1998 for less than the price of a new pickup. Parts are the hard part. When the gearbox cracked in 2021, she found a replacement on a salvage yard's website and drove eleven hours to pick it up.
“You don't fix a boat like this. You negotiate with it.”
It's important to note that maintaining aging infrastructure presents unique challenges. From sourcing replacement parts to navigating budget constraints, local governments must balance competing priorities while ensuring safety and reliability. Ultimately, investing in dependable transportation is essential for the long-term well-being of rural communities.
The county has twice proposed replacing the ferry with a larger boat that would need a crew of three. Both times the plan stalled at the dock, which is too shallow at low tide for anything with a deeper hull. Dredging would cost more than the boat. So the old ferry keeps running, and Anand keeps a notebook of every repair on the shelf above the wheel.
As the town looks toward the future, questions remain about how best to preserve this beloved service. Residents hope that a solution can be found that honors the ferry's history while meeting the needs of a growing community.
For step-by-step setup on social sites, see how to filter AI-generated posts on X, LinkedIn, and Reddit.
How to try Botfilter
Botfilter is in development, and its Chrome Web Store listing is coming soon. Start at the Botfilter home page, which shows the three display modes and will link to the store listing when it opens. Botfilter is free and needs no account. It sends no page text, scores, or usage anywhere, and the privacy policy describes what it stores on your computer.
Congrats on the team, Daniel. The 2019 part hit home for me.