Frequently Asked Questions
Hupmapper measures how original a piece of writing is by scoring how far it goes beyond what baseline AI would produce for the same task. You get an Originality Intensity (OI) score from 0 to 300.
The Originality Intensity (OI) scores from Hupmapper range from 0 to 300. Any score above 0 means there is something in the document that goes beyond Common AI – there’s originality in there. The higher the score, the more the content has diverged from Common AI, and the more originality there is throughout the document. Scores should be interpreted continuously, the difference between going from an OI of 50 to 55 indicates you have successfully distanced your content a bit more from Common AI. But, going from a 50 to 150 indicates your content has made a big jump in its distance from Common AI. Scores approaching the 200 to 300 range indicates your entire document is consistently and substantially diverging from Common AI.
We provide an overall Originality Intensity (OI) score that captures how much your entire document diverges from Common AI. But, if you click on: “Click here for Originality Breakdown” or “What does this mean?”, you will find we provide OI scores on the same 0 to 300 scale for smaller sub-sections of your document – we call them Originality Zones. This allows you to identify where the originality lives in your document, where the peaks and valleys are of OI throughout your document.
Common AI or Baseline AI refers to the typical responses you get from AI with basic prompting. Common AI is the typical responses that AI can produce easily without a human iterating with AI to add their insight, judgment, and specific goals. AI has been trained on much of human knowledge, which means even with basic prompting, AI can perform a vast array of tasks competently. But, your collaboration with AI is what will push it to generate something beyond its baseline, beyond the most common responses anyone could get from basic prompting. Even as AI surpasses increasingly far reaching and complex benchmarks – it still brings none of your unique experience and insight to the table.
Some may label Common AI responses AI slop. If one defines AI slop as the content that anyone could get from AI without putting any thought into their collaboration with AI, then yes, Common AI is similar to AI slop.
Any OI score above 0 indicates you brought something to the content that AI would not typically generate. You brought your insight, experience, unique background, judgment, local context, current goals, hard fought persistence, motivation, determined goal achievement, flexibility in problem solving, and much more. You brought all this using any tools at your disposal, including AI. We do not detect whether or not you used AI. We measure how much you used your insight and motivation to create something that goes beyond what AI commonly produces.
If you are collaborating with AI to generate content, even without writing a single word, your contribution of insight, judgment, persistence, determined goal achievement will be reflected in your prompting behavior, and subsequent AI output. Depending on your goals, you can push AI to integrate concepts nobody else has, and express ideas in ways nobody has before. I represents both your ideas and how you express those ideas. OI captures these multi-faceted characteristics of originality – the many ways you can go beyond Common AI output.
There are many ways to diverge from Common AI, so our approach captures the multi-faceted characteristics of originality. There are many ways for your content to diverge from Common AI, including ideas, concepts, themes, analogies, metaphor, expressive style, and more. We capture these characteristics by creating many metrics from the way LLMs encode and decode what is input, including how LLMs represent your text in a universal idea space – called an embedding. An embedding locates your text, in all its nuance and complexity of ideas and expression, in a universal idea space. Like a set of longitude and latitude coordinates for a geographical location. We can then quantify how far your text is from other texts in this idea space. Does it sit close to Common AI ideas and expressions, or does it diverge and truly expand the idea space? If it pushes into new territory, it expands the idea space and receives a higher OI score. We combine the multi-faceted originality characteristics to ensure that no single approach to generating originality dominates the score.
Consider this analogy that helps explain the multi-faceted approach we take to measuring originality. Imagine a city skyline. Now imagine 4 cars driving into the city from the North, South, East, and West. Each will see the same tallest buildings tower above the rest, but each will also have a unique perspective, unique angles, different orientations, see different buildings – we capture those unique characteristics of originality and combine them into an OI score.
First, a reminder that a low score is not necessarily bad, it just means that the content did not go beyond Common AI. And for past published content, AI has likely seen it or something like it, so it will be harder to go beyond Common AI for that content.
AI contains the sum of most human knowledge to date. The Gettysburg Address gets an OI score of 0. This does not mean it isn’t valuable, it just means the AI has seen the text before, so it can reproduce it without any original insight on your part – it is Common AI. Now if instead, you asked AI to write a new piece that mixes the writing styles of the Gettysburg Address and Dolly Parton’s lyrics and focus the piece on the argument of having cats versus dogs as pets (as we did) – you’ve added something original and the OI score will approach 200 or higher.
Another reason high-quality published content can score lower than even some unpublished content is that AI is trained to reproduce high quality, polished content, especially at post-training. In other words, AI mimics high quality published content in style and content. This does not mean you cannot go beyond Common AI while also writing with polish, but it does mean your ideas, style, insight, unique context, will need to diverge in some way.
We still have a unique perspective, experience, insight, motivation, and context that AI does not possess. That is what allows us to go beyond Common AI and should mean we always will. OI empowers you to identify where it is and amplify it.
There are topics and ways of expressing ideas that are more represented in AI because it has more exposure to it and is trained to reproduce it. Because the frontier AI models do not release their training datasets or training protocols, one cannot know which topics or methods of expression will look most like Common AI. This is another reason Originality Intensity (OI) scoring is such a powerful tool, as the score will tell you how much the topic(s), ideas, and ways of expressing those ideas are like Common AI. That said, it seems clear that baseline AI often reproduces the polish and style you expect to see in high-quality published content. This may mean that if you are writing about topics that have been commonly written about in the near and more distant past, it will be harder to diverge from Common AI. It also means that the more your style matches the polish of AI, the harder it will be to diverge from Common AI without incorporating new insight and ideas.
There are many ways to diverge from Common AI, so our approach captures the multi-faceted characteristics of originality - including ideas, concepts, themes, analogies, metaphor, expressive style, and more. We combine the multi-faceted characteristics of originality to ensure that no single approach to generating originality dominates the score.
AI detectors are designed to detect the unique patterns and styles of human writing. This means they will miss how you can bring your goals, insight, and unique context to AI prompting – thereby bringing originality to the content, even if AI is doing the writing. You push AI beyond its baseline, you collaborate with AI to generate new ideas, express ideas in new ways, integrate concepts in original ways, and can find many other ways to be original. Even if every word is written by AI, you collaborated with AI to push it into new territory. AI detectors label original work that human may have produced collaboratively with AI - and goes well beyond Common AI - as “AI, not human” and reject its value. We label that “Originality” and measure how much of it you bring using any tools at your disposal.
AI detectors police whether humans or AI produced something. They police authenticity and value. We do not. We empower people to make visible the many ways that they can, with or without AI, expand into new territory, unexplored by Common AI. We don’t judge the content’s value, regardless of whether the OI score is a 0, 20, 100, 200 or 300. We leave the judgment of a content’s value where it should be – with humans. With the humans who have the context and judgment that the AI cannot fully possess.
No. Hupmapper is tool agnostic. You can use AI heavily or not at all, because the score only reflects how far the finished work goes beyond what baseline AI would have produced.
Watermarking and provenance tools try to record whether AI was involved. Hupmapper answers a different question: how original the result is. It measures the value added on top of baseline AI, not who or what was in the room.
AI slop can mean different things to people. If one defines AI slop as the content that anyone could get from AI without putting any thought into their collaboration with AI, then yes, Common AI is similar to AI slop. And OI measure how far a piece of content diverges from Common AI. So yes, OI not only detects AI slop but gives a continuous score for how far a piece of content goes beyond this form of AI slop.
Yes. OI provides a way to measure how any piece of content diverges from Common AI. OI measures this divergence regardless of whether it was generated by a human, a human collaborating with AI, AI agents, or low vs. high effort AI models. This is another powerful use of OI scoring - it can tell you whether you need to use the low effort (less expensive) or high effort versions of AI models to help you ensure your content has originality in it. OI can tell you which agent or set of agents are diverging from Common AI the most.
Most of the time, a response from AI that came from basic prompting will produce a low score, but not always 0 – usually OI will be below 50. This is because basic prompting most often produces Common AI responses, but not all the responses from all AI models at all effort levels are exactly the same, so some will slightly diverge from the most Common AI responses, and consequently receive an OI score slightly above 0. It could be the choice of topic, or task, or AI model version, or just a very rare random generation. It is possible, but extremely rare, to get AI responses from basic prompting above 150. This is because millions of people are using AI every day, so at this scale, it is still possible to get a moderately rare response from AI with basic prompting, just very infrequently. This is another use for OI, it can tell you whether you happened to get a rare response from AI, even with basic prompting. Without OI scoring, this would remain invisible.
Our team, including two of our co-founders, consists of expert scientists, statisticians, and AI engineers who have spent years validating our approach. One example of our validation approach is to use openly available datasets from the peer-reviewed research in top scientific journals (e.g., Nature). We did not take part in the published work, and because the data is open, anyone can replicate our analysis. Soon, we will post a paper covering this part of our external validation approach. We also have peer-reviewed published research, as well as other papers available that explain and validate our approach to measuring OI. We will add links to them in this section as they become available.
Link to preprint: The link between diverse words and original ideas is weakening in the AI-era college admissions
The most important way to improve your Originality Intensity (OI) score is to use your own perspective, judgment, and goals and ensure they are reflected in your content. Your unique context is something AI does not possess and can push your content beyond Common AI. There is also recent research showing that starting with your ideas first, before engaging with AI can lead to higher originality in the final product.
Link to Adam’s recent conference paper, https://nam13.safelinks.protection.outlook.com/?url=https%3A%2F%2Fosf.io%2Fpreprints%2Fpsyarxiv%2Fjsz58_v6&data=05%7C02%7Cerich%40hupside.com%7C9f291d56d31d446d457408df1298392d%7C6842ff8dba8f4c119505a3a6734c4182%7C0%7C0%7C639250117132494468%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=MlN%2BzEqJRzK8DlEJK1OD9PARanw%2BJhnrHjhtCNjm1sQ%3D&reserved=0
We use submitted content to render the Originality Intensity score. After that we use it for only two purposes: to tie the content to the scoring for sharing functionality and to allow us to make sure our model's scoring is accurate. Your content is your own. We will not share it, use it, or render it, in any other way. For more information, look at our Terms of Service and Privacy terms.
They might, OI tells you how much your content diverges from Common AI, so when Common AI changes, your OI score should change along with it. But, OI scores for the same content will not change often and not as substantially as you might think. While AI is increasingly surpassing benchmarks with close-ended answers, its ability to generate distinct content in open-ended tasks that moves away from Common AI has not changed much over time. Research suggests that AI may actually be getting worse, not better, at diverging from Common AI for open-ended tasks in particular (References below). There are many reasons for this, but one of them is that AI needs to serve the most people across the most tasks, all while be accurate and competent. In particular, during post-training, AI is getting feedback from humans on which content is the best and most accurate. This necessitates the AI models essentially head toward the middle of the road, that is, the road most traveled. Asking AI directly to generate original content runs against the majority of its training.
This does not mean one cannot derive original content from AI. Original insight and original prompting approaches certainly can, but at that point, a human has made a contribution and moved beyond Common AI. That is what OI measures. References:
- Moon, Kibum, Kostadin Kushlev, Andrew Bank, Benjamin Lira, Indre Viskontas, James C. Kaufman, Dan R. Johnson, Angela L. Duckworth, and Adam E. Green. "The Link Between Diverse Words and Original Ideas Is Weakening in the AI-Era College Admissions." (2026).
- Patel, Nirav, Josiah Crossman, Eva Aggarwal, and Emily Wenger. "Are LLMs becoming similarly creative? Evidence from three years of models." arXiv preprint arXiv:2608.19437 (2026).
- Sourati, Zhivar, Farzan Karimi-Malekabadi, Meltem Ozcan, Colin McDaniel, Alireza Ziabari, Jackson Trager, Ala N. Tak, Meng Chen, Fred Morstatter, and Morteza Dehghani. "The shrinking landscape of linguistic diversity in the age of large language models." Nature Human Behaviour (2026): 1-18.
