Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Not exactly. The headline appears to refer to a 2026 preprint titled The Impact of AI-Generated Text on the Internet, by Jonas Dolezal, Sawood Alam, Mark Graham and Maty Bohacek. The paper’s public metadata does not identify the authors as Google researchers. Its results support concern about declining textual variety and increasingly positive, formulaic language, but they do not prove that AI has already made the web less accurate or that a global “dead internet” has arrived.
What the headline is actually referring to
The closest matching study is a 2026 preprint, The Impact of AI-Generated Text on the Internet. The authors sampled websites archived from 2022 through 2025 and used a state-of-the-art AI text detector to classify content. The paper estimates that roughly 35% of newly published websites were AI-generated or AI-assisted by mid-2025.
That is an estimate from archived sites and an automated detector, not a census of every page currently online. The paper metadata also does not establish a Google affiliation, so describing it as a paper “by Google researchers” is unconfirmed.
What the study found
| Question | Reported result | How to interpret it |
|---|---|---|
| How common was detected AI text? | About 35% of newly published websites were classified as AI-generated or AI-assisted by mid-2025. | A measured estimate for the study’s archived-web sample, not the percentage of the entire live web. |
| Did content become more alike? | More AI-generated text was negatively correlated with semantic diversity. | The language used across sites tended to cover a narrower range of meanings or ideas. |
| Did tone change? | AI-generated text was positively correlated with the prevalence of positive sentiment. | A more uniformly upbeat tone was associated with higher AI-text prevalence. |
| Did factual accuracy fall? | The study found no statistically significant evidence of a decline in factual accuracy. | It does not support saying that AI-written web content is already less truthful overall. |
| Did writing style become less varied? | The study found no statistically significant decline in stylistic diversity. | Semantic sameness and stylistic sameness are different measurements; only the former showed the reported negative relationship. |
The perception gap
A survey of U.S. adults found that a majority believed all four hypothesized negative effects of AI-generated text. People who used AI more frequently tended to believe those effects less strongly. This measures public attitudes, not whether either group was objectively correct.
#1 Best Overall
Does this prove the “dead internet” theory?
No. The preprint measures a rising share of detected AI-assisted or AI-generated publishing and associations with semantic diversity and sentiment. It does not establish that most online activity is generated by bots, that humans have stopped creating content, or that the web has entered a single, coordinated collapse.
A separate short paper from the ACM Web Conference 2026, Retrieval Collapses When AI Pollutes the Web, describes a structural risk rather than documenting a completed collapse. If search engines or retrieval-augmented generation systems use model-produced text as evidence, later systems may learn from synthetic material instead of primary reporting. Repeated cycles could amplify errors, omissions or uniform phrasing. The paper’s point is that this feedback loop is possible; it is not proof that a worldwide retrieval collapse has already happened.
Does AI-generated content make search results less accurate?
The available evidence gives a qualified answer. The 2026 preprint did not find statistically significant evidence that increased AI text reduced factual accuracy. That finding rules out a simple claim that more detected AI content has already made the whole web demonstrably less accurate.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteAccuracy can still be threatened indirectly. Retrieval systems may copy or rank synthetic pages, and a model may treat an earlier model’s output as if it were an independent source. The risk is greatest when duplicated text displaces original documents, when citations are missing, or when a confident answer is generated from a narrow set of mutually copied pages. Those are failure modes to monitor, not measurements showing that every AI answer is unreliable.
Rank #3
What Google says about traffic and AI Overviews
On 6 August 2025, Google Search vice president Liz Reid wrote that “total organic click volume from Google Search to websites has been relatively stable year-over-year” and that Google was sending “slightly more quality clicks to websites than a year ago.” She also said average click quality had increased.
Google says people are asking longer and more complex questions, some quick-answer searches naturally produce fewer follow-on clicks, and traffic is shifting among sites. The company also says its AI responses include prominent links, visible source citations and in-line attribution.
These are Google’s internal assessments, not an independent industry-wide measurement. They can be true alongside losses for particular publishers: aggregate clicks may remain stable while clicks move away from some sites, topics or business models. A stable total therefore does not show that every publisher is unharmed, and it does not by itself prove that AI Overviews are “stealing” traffic.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →How often do AI answers cite what they read?
An independent Social Science Research Council working paper examined approximately 14,000 real-world LMArena conversation logs. It reported that 34% of Google Gemini responses and 24% of OpenAI GPT-4o responses were generated without explicitly fetching online content.
For Gemini, 92% of answers provided no clickable citation source. The researchers call the difference between relevant URLs read and URLs actually cited an “attribution gap.” Across the models they studied, citation efficiency ranged from 0.19 to 0.45. In practical terms, a system can consult more web material than it exposes to the user, making it difficult to see which pages influenced an answer or to credit the original publisher.
The authors recommend standardized telemetry plus disclosure of search traces and citation logs. Their results concern the logged conversations and models they analyzed; they are not a universal rate for every AI product or query.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How the claims fit together
| Claim | Evidence type | What it can and cannot establish |
|---|---|---|
| AI text is becoming common on newly published sites. | Archived-web sample and detector estimate. | Supports a substantial presence in the sample; not a live-web census. |
| More AI text is associated with less semantic variety and more positive tone. | Correlation in the preprint. | Shows an association, not that AI alone caused every change. |
| AI will contaminate future retrieval. | ACM paper’s modeled or structural risk. | Explains a feedback mechanism; does not demonstrate a completed global collapse. |
| Google Search traffic is stable overall. | Google’s own traffic and click-quality assessment. | Describes Google’s aggregate view; does not resolve effects on individual publishers. |
| AI systems often fail to show sources. | SSRC analysis of about 14,000 LMArena logs. | Documents an attribution gap in the studied systems and conversations, not all AI answers. |
What readers and publishers should take from this
For readers
- Treat a fluent AI answer as a starting point, not automatic evidence.
- Open cited sources when a claim matters, and look for an original report, document or dataset rather than a chain of copied summaries.
- Be cautious when several pages use nearly identical wording or make unusually broad claims without links.
For publishers
- Original reporting, primary documents, transparent methods and distinctive analysis are harder to replace with a generic generated summary.
- Track changes in impressions, click-through rate, referrals and conversions separately; an industry-wide traffic claim may conceal large differences among sites.
- Make authorship, sourcing and update dates clear so both readers and retrieval systems can distinguish primary work from derivative text.
For search and AI-system designers
- Separate model-generated material from independently verified source content during retrieval and ranking.
- Expose which pages were consulted and which were cited, rather than presenting a single answer with opaque provenance.
- Measure semantic diversity, factual accuracy, citation coverage and source quality separately; one metric cannot stand in for all of them.
Bottom line
The evidence supports a narrower conclusion than the headline. AI-assisted publishing appears to be widespread in the studied sample and is associated with less semantic variety and a more uniformly positive tone. A retrieval feedback loop and a substantial citation gap are credible risks. But the cited preprint is not confirmed as Google-authored, it did not show a statistically significant fall in factual accuracy or stylistic diversity, and neither it nor the ACM paper proves that the internet has already been ruined.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
