The art of the bluff: writing fake Wikipedia summaries

Reverse Bluff is the hardest mode in Bluffpedia. It's also the most fun, once you figure it out.

The concept is simple: instead of spotting fakes, you write one. The game gives you a topic, and you write a Wikipedia-style summary. Then an AI evaluates how convincing your fake is. The more Wikipedia-like your writing, the higher your score.

Most players score terribly on their first few attempts. They write something that sounds vaguely informative and wonder why the AI gives them a 2 out of 10. The problem is usually the same: they're writing what they think Wikipedia sounds like, rather than how Wikipedia actually sounds.

This guide will fix that.

The Wikipedia formula

Wikipedia has one of the most distinctive writing styles on the internet, and most of it comes down to a few specific patterns. Learn these and you're already ahead of most players.

The first sentence

This is where most fakes fall apart. Wikipedia's first sentence follows an almost rigid template:

[Subject] is/was a [category noun] [qualifying details], [additional context].

Examples from real articles:

• "Eli Whitney was an American inventor, widely known for inventing the cotton gin in 1793."
• "The Danube is the second-longest river in Europe, after the Volga."
• "Photosynthesis is a biological process used by many cellular organisms to convert light energy into chemical energy."

The first sentence always does the same job: it tells you what the thing is, categorizes it, and gives you one or two identifying facts. It's definitional, not narrative. It doesn't set a scene or create intrigue. It classifies.

Where beginners go wrong: they write first sentences that sound like the opening of a magazine article. "In the heart of the Pacific Ocean lies a small island with a remarkable history." That's not Wikipedia. Wikipedia would say: "Nauru is an island country in Micronesia, a subregion of Oceania, in the Central Pacific."

Neutral tone (really neutral)

Wikipedia's "neutral point of view" policy means the text doesn't take sides, express opinions, or use evaluative language. This is harder to maintain than you'd think. Words like "famous," "legendary," "impressive," and "controversial" all carry implicit judgments. Wikipedia avoids them, or qualifies them with attribution.

Bad: "The cathedral's breathtaking architecture attracts millions of visitors." Wikipedia: "The cathedral receives approximately 12 million visitors annually."

See the difference? The second version lets the number do the talking. No adjectives needed.

Specificity over generality

Real Wikipedia summaries are packed with specific details. Dates, measurements, population numbers, coordinates, the names of related people and places. AI-generated fakes (and human-written fakes) tend to lean on vague phrasing because inventing convincing specifics is hard.

This is actually your secret weapon in Reverse Bluff. If you load your fake with plausible specifics, it immediately sounds more convincing. The trick is making those specifics plausible. More on that in a moment.

Before and after: fixing bad fakes

Let's look at some examples. These are based on real submissions from Reverse Bluff players, rewritten to illustrate common mistakes.

Example 1: A historical figure

Bad fake: "Johann Kessler was a notable German composer who lived during the Romantic period. He was known for his beautiful symphonies and his influence on later musicians. His work has been performed in concert halls around the world."

What's wrong here? Everything is vague. "Notable." "Beautiful." "The Romantic period" (when exactly?). "Concert halls around the world" (which ones?). This reads like a student padding a book report.

Convincing fake: "Johann Kessler (1811-1874) was a German Romantic composer and conductor, active primarily in Leipzig and Dresden. He composed four symphonies, two operas, and numerous chamber works. His Symphony No. 3 in F minor (1847), sometimes called the Dresdener, was a regular fixture of the Gewandhaus Orchestra's repertoire during the 1850s. Kessler served as Kapellmeister at the Semperoper from 1852 until his death."

This version has birth and death years, named cities, numbered works, a specific key signature, a nicknamed composition, a named orchestra, a job title, and dates of employment. None of it is real. But it all sounds real because it follows the patterns of actual Wikipedia entries about composers.

Example 2: A place

Bad fake: "Lake Vorinsee is a large lake located in the Swiss Alps. It is known for its crystal-clear waters and stunning mountain scenery. The lake has been a popular tourist destination for many years."

Convincing fake: "Lake Vorinsee is an alpine lake in the canton of Graubuenden, Switzerland, situated at an elevation of 1,847 metres (6,060 ft) above sea level. The lake has a surface area of 2.3 km2 (0.89 sq mi) and a maximum depth of 74 metres (243 ft). It is fed primarily by glacial meltwater from the Vorin Glacier and drains northward into the Hinterrhein. The area was designated a Swiss Federal Inventory of Landscapes site in 1983."

The parenthetical unit conversions are a classic Wikipedia touch that most people forget. The drainage pattern, the specific glacier name, the federal designation with a year: these are the kinds of details that make a fake feel real. Notice there's no mention of "stunning scenery" or "crystal-clear waters." Wikipedia doesn't care how pretty a lake is. It cares about the elevation and surface area.

Trait	Convincing fakes	Obviously fake
First sentence	Definitional: '[Subject] is a [category]...'	Narrative or dramatic opening
Specificity	Exact dates, numbers, names, coordinates	Vague: 'many years,' 'around the world'
Tone	Neutral, dry, factual	Enthusiastic, evaluative, or literary
Adjectives	Few, and only factual (e.g., 'third-largest')	Subjective: 'famous,' 'stunning,' 'remarkable'
Structure	Classification then details, often with parentheticals	Story arc with beginning, middle, end
Details	Internally consistent, plausible for the subject type	Generic, could apply to almost anything
Length	Matches typical Wikipedia summary length (4-8 sentences)	Too short (1-2 sentences) or too long

Advanced techniques

Once you've got the basics down, these techniques will push your scores higher.

Steal structure from real articles

Before playing Reverse Bluff, spend five minutes browsing Wikipedia articles about subjects similar to your target. If you're writing about a fictional city, read three real Wikipedia city articles. Notice the order of information: definition, geography, demographics, history, economy. Copy that structure with invented details.

The evaluator AI was trained on Wikipedia text. It knows what order information usually appears in for a given subject type. Matching that order signals authenticity.

Use Wikipedia's verbal tics

Every writing style has its quirks. Wikipedia has a bunch of them, and sprinkling them into your fake adds authenticity.

Parenthetical alternate names: "Kyoto (formerly known as Heian-kyo)" Inline unit conversions: "482 kilometres (300 mi)" "As of" phrasing: "As of the 2020 census, the population was 34,219" Disambiguation cues: "Not to be confused with..." (though this usually appears at the top of articles, not in summaries) Attribution hedging: "According to historian James Worth..." "Known as" or "referred to as" constructions for nicknames

Make your numbers plausible

This is where most players get caught. If you're writing about a city, what's a reasonable population? A made-up German city probably shouldn't have 5 million people (that would make it bigger than Berlin). A small Swiss lake probably isn't 200 km2 (that would be bigger than Lake Zurich).

You don't need to be exact. You just need to be in the right ballpark. When in doubt, go smaller. A population of 43,780 sounds more plausible than 500,000 for a city you've never heard of. An elevation of 1,847 meters sounds right for an Alpine lake. These small calibrations make a huge difference.

Include one slightly boring detail

Real Wikipedia articles often contain at least one piece of information that's technically accurate but not particularly interesting. The administrative district a town belongs to. The geological classification of a rock formation. The ISO standard that governs a technical process.

Fakes tend to be too interesting because the writer is trying to make them convincing, and "convincing" gets confused with "engaging." But Wikipedia isn't trying to be engaging. It's trying to be comprehensive. Include a dull fact. It makes everything else around it more believable.

Why this skill actually matters

We pitch Reverse Bluff as a fun creative challenge, and it is. But the skill you're developing has real-world value that goes beyond the game.

When you learn to write convincing Wikipedia-style text, you're learning to understand how AI generates text. Large language models produce text by predicting the most likely next word based on patterns in their training data. When you write a fake that scores well, you've essentially reverse-engineered that process. You've figured out what patterns the model expects, and you've produced text that matches them.

That understanding cuts both ways. Once you know how to produce convincing AI-style text, you also get better at detecting it. You start to notice the seams. You recognize when a piece of text is following a statistical pattern rather than conveying genuine knowledge. The best Bluffpedia players are the ones who've played a lot of Reverse Bluff, and we don't think that's a coincidence.

In a world where AI-generated text is everywhere, from customer service chats to news articles to student essays, understanding how that text is constructed is a form of literacy. Reverse Bluff teaches it through play, which is the best way to learn anything.

Go practice

Your first few Reverse Bluff attempts will probably be rough. That's normal. Keep the comparison table from this article open in another tab, write your fake, check it against the "convincing" column, and revise before submitting.

After about 20 rounds, something clicks. You stop thinking about the rules and start feeling what Wikipedia sounds like. Your fakes get better. Your detection skills in other modes improve too. And you'll never read a Wikipedia article the same way again, because now you notice all the little structural patterns that make it Wikipedia.

Good luck fooling the AI. It's been reading a lot of Wikipedia, so it won't be easy.