What does text copied from Gemini actually look like?
Gemini writes its answers in Markdown: ## for headings, **bold** for emphasis, - for bullet lists, and full | Col A | Col B | tables, which Gemini reaches for more readily than most chat models when a comparison is involved. Gemini's own interface renders all of that into formatted text, so it looks clean there. The trouble starts when you copy the rendered result into a destination that doesn't render markdown the same way — the raw symbols come along as literal characters, and a wide table can wrap or truncate badly in a narrow field.
Separate from markdown, Gemini's prose leans toward the same "smart" typography documented across current large language models — em dashes, curly quotes, ellipses. See why AI text is full of em dashes for the data behind that pattern.
Where the hidden characters actually come from
It's tempting to assume every invisible character in AI output is the model "marking" its text. For Gemini, that's not accurate. The one confirmed case of a chatbot inserting a trackable hidden character was OpenAI's o3 and o4-mini briefly emitting narrow no-break spaces (U+202F) in April 2025 — OpenAI called it a quirk of large-scale reinforcement learning, not a watermark, and it stopped within days. There is no equivalent documented episode for Gemini.
What does show up in Gemini pastes is ordinary paste-pipeline noise: non-breaking spaces from the chat UI's HTML, curly quotes and em dashes from the model's writing style, and occasionally a byte-order mark or zero-width character introduced by whatever app you pasted into next — Word, Google Docs, Notion, a CMS editor. See why text pastes weird for the full list of culprits, or the invisible Unicode characters list to see exactly what each one is.
SynthID-Text: Gemini's real watermark, and why cleaning can't remove it
Google DeepMind has run SynthID-Text in Gemini since 2024, live-testing it on roughly 20 million real Gemini responses before publishing the method in Nature in October 2024. Instead of inserting a character, it uses "tournament sampling": as Gemini generates each word, a secret key scores competing candidates and nudges the choice toward a winner, leaving a pattern invisible to a reader but testable by whoever holds the key. See SynthID-Text explained for the full mechanism.
No Unicode characters are added, so there is nothing for a character cleaner to find or strip. Anthropic adopted the same tournament-sampling approach for Claude models launched on or after August 2, 2026 — see cleaning Claude text for that side of the comparison.
Honesty note: this tool removes character-level artifacts only. It is not, and does not claim to be, a way to defeat AI detection or strip a statistical watermark — see how AI detectors actually work for why word-choice patterns and hidden characters are separate problems.
Gemini vs. ChatGPT vs. Claude: character quirks and watermarks
| Model | Documented character-level quirk | Statistical watermark | Unicode tag characters sanitized? |
|---|---|---|---|
| Gemini | None documented | SynthID-Text, since 2024 | No (FireTail, Sept. 2025) |
| ChatGPT (o3 / o4-mini) | Narrow no-break space (U+202F), April 2025 — fixed within days | None announced | Yes (FireTail, Sept. 2025) |
| Claude (Aug. 2, 2026+) | None documented | SynthID-Text-style, since Aug. 2026 | Yes (FireTail, Sept. 2025) |
Sources: Google DeepMind, Nature 634:818–823 (Oct. 2024); Anthropic via TechCrunch (Aug. 11, 2026); Rumi (April 2025); FireTail (Sept. 2025). See ASCII smuggling explained for what the tag-character column means and why it matters for pasted text fed back into an AI tool.
Cleaning Gemini output before you paste it somewhere else
Paste Gemini's response into the cleaner above. The AI Clean preset (default) strips invisible and non-breaking characters and standardizes typography while leaving markdown symbols intact — the right choice for most destinations. If you're pasting into a field that shows raw text instead of rendering markdown, turn on Strip Markdown in Advanced Controls — this matters more for Gemini than most models, since its table syntax turns into a wall of pipes and dashes if left unrendered. If you'd rather keep Gemini's em dashes and curly quotes, use Keep Typography instead. For code, CSV, or any system that only accepts keyboard characters, use Plain ASCII. Full destination-specific notes live on pasting AI text into Word and Google Docs.