Is AI content bad for SEO? What CNET and Gannett actually show

Is AI content bad for SEO? The evidence says no, but the four biggest AI publishing scandals of 2023, at CNET, Sports Illustrated, Gannett, and G/O Media, make the fear feel justified even when the data doesn't support it. Each one is worth walking through in detail, because each one is routinely cited as proof that AI content tanks your Google rankings, and none of them actually was.

What sank those four newsrooms was publishing AI output nobody checked, not the AI itself. CNET corrected 41 of 77 finance articles. Sports Illustrated ran product reviews under invented author personas. Gannett and G/O Media both pulled AI tools after factual errors went viral. In every case, the consequence was an editorial and reputational one: corrections, paused programs, fired executives. Not one came with a Google statement or a documented ranking action naming AI authorship as the cause. This post walks through the four cases, then what Google's own policy and rater guidelines actually say, then the two large-scale studies that separate authorship from penalty. If you want the full 600,000-page dataset behind the "no penalty" finding, our deep-dive covers it in detail; if you want the short, decision-ready version with a pre-publish checklist, is AI generated content bad for SEO walks through Google's guidance in five minutes. This piece is about the specific evidence: what happened at four real newsrooms, and what it does and doesn't prove.

Is AI content bad for SEO? The short answer

No, not for being written with AI. Google's spam policy and its rater guidelines both judge pages on originality and effort, not on the tool that drafted them, and the largest public study on the question, Ahrefs' analysis of 600,000 pages, found a correlation of 0.011 between a page's AI-content share and its ranking position. That's statistically indistinguishable from zero.

What the 2023 case studies actually show is a different risk entirely: publishing unverified AI output damages trust and invites correction, whether or not it ever touches a ranking. CNET, Sports Illustrated, Gannett, and G/O Media all learned that the hard way, and all four learned it through editorial fallout, not an algorithm update. The rest of this post separates the two risks, because conflating them is exactly what keeps a defensible AI workflow stuck in another quarter of evaluation.

What Google's guidelines actually say about AI-assisted content

Google has published three layers of guidance on this, and they converge on the same point: authorship isn't the variable, effort and value are.

The spam policy: judged "no matter how it's created"

Google's spam policies documentation defines scaled content abuse as "many pages generated for the primary purpose of manipulating search rankings and not helping users," targeting "large amounts of unoriginal content that provides little to no value to users, no matter how it's created." That last clause removes the production method from the test entirely. A human content farm and an AI content farm trip the same wire, and so does a single writer, human or automated, publishing thin pages at volume. The two things the policy actually names are purpose and value.

Google's dedicated AI content guidance (Feb 2023, updated Dec 2025)

Google addressed AI content directly for the first time in its February 2023 guidance: "Rewarding high-quality content, however it is produced, is key to what we do." That guidance has since been folded into a dedicated page on generative AI content, last updated December 10, 2025, which states that "using generative AI tools... to generate many pages without adding value for users may violate" the scaled content abuse policy, and recommends disclosing how content was produced so readers have context. Danny Sullivan, Google's Search Liaison, put the underlying logic even more bluntly: "we don't really care how you're doing this scaled content, whether it's AI, automation, or human beings. It's going to be an issue."

The Helpful Content System and E-E-A-T in an AI content world

Google folded its standalone Helpful Content System into the broader ranking systems in 2024, but the standard it enforced didn't go away: pages need to demonstrate real experience, expertise, authoritativeness, and trust, the qualities Google's Search Quality Rater Guidelines use to judge whether a page is worth surfacing at all.

What the Search Quality Rater Guidelines actually instruct raters to check

Raters aren't asked whether a model wrote the page. Google's Search Quality Rater Guidelines instruct raters to assign the Lowest quality rating to pages whose main content is "copied, paraphrased, embedded, auto or AI generated... with little to no effort, little to no originality, and little to no added value". Read that condition again: "little to no effort" is an editorial standard, not a detection standard. It's the same test a competent human editor already applies to a freelancer's draft.

Why "Trust" is the E-E-A-T factor unreviewed AI content fails first

Of the four E-E-A-T pillars, Trust is the one the guidelines call load-bearing, because an untrustworthy page scores low no matter how experienced, expert, or authoritative it otherwise looks. Unreviewed AI content fails Trust first and most often, for a mechanical reason: a language model predicts plausible text, not verified text, so a fabricated statistic and a real one come out of the same process looking identical. Nothing in the drafting step catches that. Only a review step does, which is why the editorial review process for AI content treats fact-checking and link verification as a separate, independent pass rather than something the writer grades on its own draft.

Case evidence: is AI content bad for SEO, or was it never reviewed?

This is the part worth sitting with, because these four cases get cited constantly as proof that AI content is radioactive for SEO, and none of them actually demonstrate that. Read them for what they do show: what happens when AI output ships without anyone checking it first.

CNET: 41 of 77 AI finance articles needed correction

CNET began quietly publishing AI-written personal-finance explainers in November 2022 under the byline "CNET Money Staff," without disclosing which articles were AI-written until Futurism flagged factual errors in January 2023. Once CNET reviewed its own archive, it corrected 41 of the 77 AI-written articles, a 53% correction rate, and paused the program that same month. CNET's editor-in-chief said some of the errors required only minor fixes, like unclear wording, but others needed substantial correction. The company said it would restart the tool once it was confident the editorial process could catch both human and AI errors before publishing, not after.

Sports Illustrated: fabricated AI bylines, an executive shakeup, and no policy violation ever cited

In November 2023, Futurism reported that Sports Illustrated had published product reviews under invented author personas, "Drew Ortiz" and "Sora Tanaka," whose headshots traced back to an AI headshot marketplace and who had no verifiable byline history anywhere else online. The content came from third-party vendor AdVon Commerce, which told parent company Arena Group the articles were "written and edited by humans" using pen names, a claim Arena Group's own sourcing contradicted. Arena Group terminated the AdVon partnership and pulled the content. Within weeks, it fired CEO Ross Levinsohn, along with COO Andrew Kraft, media president Rob Barrett, and corporate counsel Julie Fenster. The stated reason was operational efficiency and revenue, not a search penalty; there was never a Google statement tying the incident to Sports Illustrated's rankings.

Gannett and G/O Media: paused experiments after public, factual failures

Gannett, the largest US newspaper chain, deployed AI vendor LedeAI in August 2023 to auto-generate local high-school sports recaps across outlets including the Columbus Dispatch, the Louisville Courier Journal, and the Milwaukee Journal Sentinel. The copy went viral for robotic phrasing and unfilled template placeholders like "[[WINNING_TEAM_MASCOT]]" left directly in published articles, and Gannett paused the LedeAI experiment across every market using it. The same month, G/O Media's io9 published an AI-generated Star Wars chronology under the byline "Gizmodo Bot," with no advance notice to the site's editorial staff. It contained factual errors, including a mangled release-order claim, and omitted entire Disney+ series. io9's deputy editor, James Whitbrook, sent a lengthy list of corrections to G/O Media and publicly called the piece "embarrassing, unpublishable, disrespectful of both the audience and the people who work here, and a blow to our authority and integrity." The article stayed up but went through that correction pass; staff pushed back publicly, and G/O Media didn't repeat the experiment on io9.

The pattern across all four: reputational and editorial collapse, not a documented Google ranking penalty

Line the four cases up and the pattern is identical: unreviewed AI output shipped to readers, someone outside the newsroom caught the errors, and the consequence was corrections, a paused program, or a firing. Not one of the four came with a Google statement, a manual action notice, or any documented ranking movement tied specifically to AI authorship. That absence is itself evidence for the core claim in Google's own guidance: it grades the page's value and effort, not the production method. The damage in all four cases was to reader trust and editorial credibility, which is a real cost with or without a ranking effect, and it's the exact failure mode an editorial review gate exists to catch before publication, not after a viral thread forces a correction.

The real risk factors: thin output and no editorial review, not AI itself

Two large, independent datasets separate the two risks cleanly: whether AI authorship correlates with ranking, and whether unedited AI output performs as well as edited work. They answer different questions, and both matter.

The data that separates authorship from penalty: Ahrefs' 600k-page correlation

On July 7, 2025, Ahrefs researchers analyzed the top 20 ranking URLs for 100,000 keywords, 600,000 pages in total, and found the correlation between a page's AI-content percentage and its ranking position was 0.011. Correlations run from -1 to 1; 0.011 is noise. In the same dataset, 86.5% of top-ranking pages already contained some AI-generated content. If Google were penalizing AI authorship directly, that number would look nothing like this.

The data that shows unedited AI still underperforms: Semrush's 42k-post gap

Authorship not being the penalty doesn't mean unedited AI performs as well as edited content. Semrush's November 2025 analysis of 42,000 blog posts found human-written content held Position 1 about 80.5% of the time versus roughly 10% for purely AI-generated content, an 8x gap. "Another place where we draw a clear line today is editing," Semrush's Ana Camarena said of the finding. "That step is still fully human-led." A separate experiment tracked 2,000 fully AI-generated, unedited articles across 20 brand-new domains for 16 months: pages indexed fast and even earned early impressions, then the share still ranking in the top 100 fell from 28% to 3% by month three, with no meaningful recovery over the following year. That gap is an editing gap, not an authorship penalty. It's also the same gap the March 2024 core update targeted directly: Google reported 45% less low-quality, unoriginal content in results afterward, and independent tracking by Search Engine Journal found roughly 837 sites deindexed in its early weeks, the same scaled-content-abuse crackdown the spam policy language above was written to enforce. The pattern held into 2026: Google's March 2026 spam update rolled out and finished in under 20 hours, the shortest confirmed spam update in Google's dashboard history, with a broad core update following soon after.

How Git/PR-based editorial workflows reduce penalty exposure

If the actual risk is unreviewed volume, whether that risk shows up as a ranking filter or as a CNET-style correction spiral, the fix is a step that makes "unreviewed" structurally impossible. A Git-based, pull-request editorial gate does that more specifically than "have someone glance at it before publishing."

What a pull request proves that a CMS publish button doesn't

A CMS publish button is a state change: draft becomes live. It proves nothing about what happened in between, which is exactly the gap that let CNET's uncorrected errors and Sports Illustrated's fabricated bylines reach readers before anyone outside the company caught them. A pull request is a record: it shows the exact diff a reviewer looked at, who approved it, and when, tied permanently to that version of the page. That record is also what AI content governance increasingly requires you to produce, separate from any ranking question, as EU disclosure rules and enterprise buyers both start treating an unreviewed AI post as a trust problem on its own terms. Nothing goes live until a named person says yes, which removes the volume risk at the source instead of trying to catch it after a post has already reached readers.

A pre-publish checklist for AI-assisted posts

Turn the four case studies and Google's own guidance into a single pre-publish checklist:

None of these are exotic. They're what a careful editor already does to any draft, and a growth channel like SEO for SaaS depends on that discipline holding at volume, not just on the first few posts. The difference between a checklist on a whiteboard and one that actually runs on every post is whether something blocks the merge when a box goes unchecked.

This is the gate Lyra is built around. She drafts in your blog's existing voice, fact-checks every claim against a current source, verifies every link resolves, and keeps a named byline on the page, closer to the review process CNET and Sports Illustrated didn't have than to the one-click publish that got G/O Media into trouble. Then she opens a pull request and tags you. Nothing auto-publishes; you read the diff and merge, or send it back. Unlike a black-box, one-click generator, that review step is the whole point, which is the same distinction our Content at Scale alternative breakdown walks through in more detail.

None of the 2023 AI publishing scandals were ever tied to a documented Google ranking penalty. They were editorial failures from shipping unreviewed output, which is exactly what a PR review gate is built to catch before it reaches a reader.

Try Lyra → · Talk to the founder