What is trust in search in 2026?
Search is becoming increasingly biased. Google has decided it’s models are the primary source of truth before humans even have the illusion of choice. Yes, illusion. Your search isn’t getting an unfiltered view of the web, the search results from every major engine is filtered according to their algorithms. The web wasn’t mean to be held in the hands of the few, it was meant as an equalizer for information.
“Believe half of what you see and none of what you hear” - said by Edgar Allan Poe, apparently, according to Google.
That quote you just read is exactly why we need new trust signals. Search engines will tell you what they think is true, not whether it’s true.

Google’s AI overview confidently puts the words in Poe’s mouth. The interesting thing isn’t that Google is hiding the chain of trust — they’re not. The first source
the overview cites is a Seacoast Online column from 2014 by a real columnist named Jerry. (Go ahead, turn off the adblocker for a second and reload it.) Jerry seems like a perfectly real person. The column is real. The attribution to Poe is the part that’s in question.
What’s more interesting is what happens next in Google’s own citation chain.
So Google cited the right reference. And then summarized it wrong. The failure isn’t that the chain of trust is hidden — it’s that the summarization layer picks the
loudest claim, not the best-supported one. Once that summary lands above the fold, the citation chain underneath stops mattering…the reader’s already moved on.
Did Poe say this? Sure, probably. Was he the creator? Almost certainly not. Does it matter for a quote about trust? That depends on your version of the truth, I guess.
For a quote, fine. For other things, it’s a real problem. Google’s AI overview infamously told users to add non-toxic glue to pizza sauce so the cheese wouldn’t slide off, recommended eating one small rock a day for minerals, and claimed geologists endorsed the rock-eating (The Conversation has a good roundup). The rock advice was scraped from an Onion article. The glue advice was scraped from an eleven-year-old Reddit comment. Real source URLs were cited. The summarizer just couldn’t tell satire and shitposting apart from instruction.
Same pattern as the Poe quote.
In hands-on comparisons against Google’s AI overview, the chat products from Anthropic, OpenAI, Perplexity, and Grok can produce more accurate answers on factual questions more often than not, because the user can follow up, immediately check if a link source exists, and dig deeper. They aren’t perfect, none of these systems are. But these systems can be more reliable, we’re just not there yet.
Audit trails are the load-bearing piece.
Google Fact Check (todo)+ open
agent manifest (json)
{
"@context": "https://opensourceseo.org/agents/v0",
"@type": "StackQuestion",
"layer": "trust-verification",
"question": "why-change-is-mandatory",
"title": "What is trust in search in 2026?",
"prompt": "Search is becoming increasingly biased. Google has decided it's models are the primary source of truth before humans even have the illusion of choice. Yes, illusion. Your search isn't getting an unfiltered view of the web, the search results from every major engine is filtered according to their algorithms. The web wasn't mean to be held in the hands of the few, it was meant as an equalizer for information. ",
"sections": [
{
"slug": "hook",
"label": "Hook",
"kind": "hook"
},
{
"slug": "why-now",
"label": "Why this question, now",
"kind": "prose"
},
{
"slug": "options",
"label": "The options",
"kind": "options"
},
{
"slug": "changes-mind",
"label": "What would change our mind",
"kind": "bullets"
},
{
"slug": "branches",
"label": "Where this leads",
"kind": "branches"
}
],
"images": [
{
"url": "https://opensourceseo.org/images/66ca0737beef.webp",
"alt": "google ai overview for edgar allen poe quote misattributed",
"filename": "google-quote-edgar-allen-poe"
}
],
"claims": [
{
"id": "9ba17af290a2",
"text": "The same AI overview lists Quote Investigator as a supporting source.",
"status": "uncertain",
"lastChecked": "2026-05-05T20:24:13.212Z",
"note": "**Cannot verify from available evidence.**\n\nThe on-page screenshot (`/images/66ca0737beef.webp`) shows Google's AI Overview citing **Seacoastonline.com +4** — meaning 4 additional sources behind a collapsed *Show more* button. Quote Investigator is **not visible** in the unexpanded screenshot. Whether QI is one of the +4 cannot be confirmed from this image alone.\n\nGoogle's AI Overview is personalized and time-varying — it changes per user, location, and date — so a third-party reviewer cannot reliably reproduce the exact citation list the author saw. WebSearch from a separate session cannot recover Google's AI Overview output for the original query.\n\nTo upgrade this to `valid`: capture and store an expanded screenshot of the same query showing all five cited domains, or have a second reviewer reproduce the AI Overview and confirm QI in the source list.\n\nNot marking `contradicted` — there is no positive evidence against the claim, only an absence of visible evidence for it."
},
{
"id": "81ffbafc4a82",
"text": "Quote Investigator documents the expression in print as early as 1831 — fourteen years before Poe is supposed to have said it, sourced from [a digitized 19th-century anthology](https://books.google.com.co/books?id=pZme1YXq_E4C&q=%22nothing+you%22).",
"status": "valid",
"lastChecked": "2026-05-05T20:24:13.212Z",
"note": "**Confirmed by primary source.**\n\nQuote Investigator's article documents the earliest known instance of the recognizable form of this expression appearing in 1831, in William Johnson Neale's novel *Cavendish: Or The Patrician at Sea*. QI quotes the Neale passage directly: *\"The rule with us is, believe nothing you hear, and but half you see.\"* The Google Books URL cited in the claim resolves to **Cavendish, Volume 1** by William Johnson Neale — the same book QI references.\n\nThe arithmetic checks: 1845 (Poe's *The System of Dr. Tarr and Prof. Fether*, the version most widely attributed to Poe) − 1831 = 14 years. Matches the claim verbatim.\n\n**Caveat worth noting in future revisions:** QI also documents *related but differently-worded* earlier appearances — a 1783 Boswell-Johnson conversation discussing believing \"one half\" of narratives, and an 1809 song lyric using \"only half what we hear.\" The recognizable form of the expression first appears in 1831; precursor variants existed earlier. The claim is accurate as written but a careful reader could press on the word *expression*. Calling it `valid` because the author is clearly referring to the recognizable form, not to any thematic ancestor."
}
],
"instructions": {
"forAgents": "Each section under `sections` is a structured component of this question. Treat `layer` + `question` as authoritative taxonomy. Cite by anchor: `/${layer}/${question}#${section}`."
}
}