Patent Families for Detecting AI-Generated Content Rose 237.5% Between 2022 and 2025, and Text Detection Grew Sixfold

Summary

Picture a marketing manager wondering whether pages written with AI can be spotted. Patent filings show that tools for detecting AI-made content are a fast-growing area of invention, especially for text.

We found 346 patent families since 2018 whose titles describe detecting AI-generated or deepfake content. Publications went from 32 in 2022 to 108 in 2025, up 237.5%. Text detection grew fastest, from 8 families to 48.

Most filings come through China, and universities or research institutes own more than half. The data does not show that any search engine uses these methods. A patent shows what was filed, not what is live in any product.

What we measured

We pulled every publication in Google Patents Public Data (patents-public-data.patents.publications) published from January 1, 2018 to September 17, 2026 that carries at least one CPC code in G06F16 (information retrieval), G06F40 (natural language processing) or G06N (AI and machine learning) and has an English abstract. That is 2,170,848 publications in 1,214,491 patent families. For each publication we kept the English title, the English abstract and up to the first 20,000 characters of the English claims. English claims exist for 488,406 publications, mostly US, European and international (WO) filings, so text matching is deeper for those offices than for Chinese, Korean or Japanese filings, which we matched on title and abstract.

To keep precision high, we matched on English titles only. A family counts when its title includes a detection word (detect, identify, distinguish, discriminate, recognize, authenticate, forensic, classify or judge) and a term for generated content, such as machine-generated, AI-generated, generated by a language model, AIGC, generated text, deepfake, deep forgery or synthetic face, image, video or text. Text and media groups come from the same titles. Matching on titles misses families that describe detection only in the abstract or claims, so the counts are a floor.

Detection filings more than tripled from 2022 to 2025

We counted families by first publication year and split text from media detection.

Media detection, mostly deepfakes, has led every year since 2020. Text detection has grown quickly since 2024, from 10 families in 2023 to 48 in 2025. By first filing year, 96 families were first filed in 2024 and 94 so far in 2025.

Most filings come through China and from universities

We counted every office where a family published and owners with academic names.

Microsoft holds 7 families, as do several Chinese universities. Intel and Bank of America hold 5 each. For text detection alone, Microsoft and IBM hold 4 each.

What this means

Detection of AI-written text is an active research and patent area, and it is growing faster than media detection. Whether any search engine or AI assistant uses such methods is not shown by this data, but the direction is clear enough to plan around. Sensible AI SEO services plan for it by keeping real people and real facts behind every page.

Use AI as a drafting aid, then add what only your business knows: real prices, local details, photos and customer questions. Pages built on first-hand facts hold up whatever a detector decides. A free AI visibility audit shows which of your pages AI assistants already trust enough to mention, and that list should guide your AI search optimization more than any detector score.

Limits

A patent shows what was filed, not what is live in any product.

Nothing in this data shows that any search engine or AI assistant uses these detection methods.

Patents usually publish about 18 months after the first filing, so families first filed in 2025 and 2026 are heavily undercounted, and 2026 publication counts cover January 1 to September 17, 2026 only.

Title-only matching favors precision over recall. A sample review found a few off-target titles (for example, synthetic images made to train other detectors), so treat counts as approximate.

Only families tagged G06F16, G06F40 or G06N are included; image-forensics patents classified only under image codes are missed.

Search-specific detection (spam or search ranking) was too rare to measure reliably and is not broken out.

Frequently asked questions

Related studies

• Patents behind AI searchFreshness Shows Up in Only 3.6% of AI Ranking Patents
• Patents behind AI searchQuery Fan-Out Patents: 44.1% Now Use Generative AI
• Patents behind AI searchSearch Patents 2026: 39.9% Now Mention Generative AI
• Patents behind AI searchChat Shopping Patents: 43.3% Now Use Generative AI