Lowest: untrustworthy pages and spam

What the document says

Untrustworthy pages (4.5)

The opening rule is harsh: Lowest for pages the rater strongly suspects are deceptive or malicious — no proof required. The logic is pragmatic: since many people refuse to use a highly suspicious page, it fails its purpose by definition. The signals: multiple or significant factual inaccuracies; inadequate information about who’s responsible (fatal on pages handling money, passwords or sensitive data — a YMYL page with zero information about the site is automatic Lowest); rock-bottom E-E-A-T or reputation; and the three forms of deception the document catalogs:

Deceptive purpose — the page appears to exist for one thing but exists for another: an “independent review” created purely to drive clicks on monetized links, a product recommendation based on personal testing that never happened.
Deceptive information about the site or creators — impersonating another site, a news-outlet facade built for political or commercial manipulation, an online store claiming a fake physical address, and two recently added items: fake AI-generated author profiles (invented photo and bio to fake human authorship) and false credentials (a creator falsely claiming to be a medical professional).
Deceptive design — ads disguised as search results or site navigation, buttons that pretend to close a pop-up but trigger a download, titles with nothing to do with the content.

Section 4.5.4 adds deliberately obstructed MC: ads that chase the scroll and won’t close, coercive interstitials, ads that push content out of sight, white-on-white text. With one explicit and important exception: paywalls and logins on trustworthy sites are not coercion — quality newspapers and magazines couldn’t exist without subscriptions.

Spammy pages (4.6)

The spam catalog mirrors Google’s Web Spam Policies, and three tactics earned their own sections in the AI era:

Expired domain abuse (4.6.3) — buying a lapsed domain (a school, government agency, charity) and repurposing its reputation for unrelated monetized content. Raters are taught to use the Internet Archive to catch the switch.
Site reputation abuse (4.6.4) — “parasite SEO”: third-party content published on an established site mainly to ride the host’s ranking signals. A medical site hosting “best casinos,” a movie-review site running “buy social media followers,” a news site publishing white-label coupons. The exceptions are precise: syndicated content, forums and genuine UGC, editorial columns, and coupons sourced directly from merchants.
Scaled content abuse (4.6.5) — many pages generated to benefit the owner, not the visitor: automated tools (generative AI or anything else) producing volume without value, scraping with synonymizing or auto-translation, stitching together other people’s content, multiple sites built to hide the scale. The rater must apply Lowest even without knowing the creation method — strong suspicion after viewing several pages is enough.

Section 4.6.6 completes the reasoning: Lowest when all or almost all MC is copied, paraphrased, embedded or reposted with little to no effort, originality and added value — even with credit to the source. But with the caveats that balance the scale: licensed/syndicated content is not copying; embeds and reposts are legitimate when value is added; paraphrase can be valuable (an expert translating public policy into plain language); and using generative AI, by itself, determines nothing — the same tools can produce Highest and Lowest work. Section 4.6.7 even teaches the forensics: search quoted sentences on Google and check original dates on the Wayback Machine.

In practice

Google has described — and now punishes in writing — the three tactics that dominated opportunistic SEO in 2023–2025. Expired domains, parasite SEO and scaled AI content each have numbered sections with human detection instructions. When a tactic earns its own section in the guidelines, its cycle is over: it has become an algorithm-training target. Anyone still selling you these tactics is selling you liability.

Fake author profiles are deception, not “E-E-A-T optimization.” The industry responded to E-E-A-T by inventing authors — stock or AI photos, generic bios, vague credentials. The document now names that literally as deceptive design, a Lowest category. A real author with verifiable credentials, or no author at all: both are defensible. An invented author, never.

The AI ruling is about process, not tools. The same section that orders Lowest for content farms says explicitly that generative AI can produce high-quality work with high effort. The rater’s question isn’t “was this made with AI?” — it’s “is there effort, originality and added value here?” A pipeline with real human curation, proprietary data and editing passes; raw volume doesn’t.

Suspicion is enough. Perhaps the most important sentence in the section: the rater assigns Lowest without proof, on the pattern alone. Sites that live at the edge of defensibility (“technically it’s not copying,” “technically the pop-up closes”) are exactly what the strong-suspicion standard catches. The safety margin isn’t legal, it’s perceptual: if it looks like a scam, it gets rated like one.

Run the plagiarism test on yourself. Take three sentences of your content and search them in quotes. If dozens of sites carry the same text and yours isn’t the identifiable original source, you’re on the wrong side of section 4.6.7 — even if you “rewrote it.”