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
In Sourcelift’s 2026 test, all 80 AI answers defined their assigned terms correctly—but that did not establish that every other claim was supported, current, or traceable. The practical test for a consequential answer is not just “Is the definition right?” It is “Can I open a credible source, and does that source support this specific claim?”
What the 80-answer test found
Tony G., founder of Sourcelift, reported the test in an article published by The AI Journal on October 5, 2026. Sourcelift ran it on September 29, 2026, asking ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews the same 16 questions about eight AI-search terms, including GEO, llms.txt, query fan-out, and E-E-A-T. Each question was asked once in English and once in Spanish, producing 80 answers.
The test used logged-in consumer apps with default settings and read Google AI Overviews on google.com. The author says answers were read in full and checked against primary sources, such as an originating paper or official documentation; claims judged unsourced, outdated, or contradicted were also flagged. These are the author’s reported findings, not an independent replication.
Recommended Free Tools
- 80 of 80 answers defined the term correctly. This is the reported result for the test’s 80-answer sample.
- 21 of 64 answers linked a primary source. This count covers ChatGPT, Claude, Gemini, and Perplexity, excluding Google AI Overviews.
- 14 of 32 English answers and 7 of 32 Spanish answers linked a primary source. These counts describe this test only; they do not establish a general language-based performance gap.
The test author cautioned: “It is a small sample, with one run per question on a single day.” Because each prompt was run once, the results are a snapshot, not a stable rate or a basis for ranking systems. AI answers can vary across runs.
#1 Best Overall
Why a correct definition can still mislead
An answer may accurately explain a term and still make unsupported claims about a number, date, recommendation, or source. Those details often carry the real decision-making weight. For example, a definition of query fan-out can be accurate even if a numerical estimate of how many searches a system performs is not documented.
In examples reported by Tony G., Google AI Overview gave a typical query fan-out estimate of 5 to 12 searches, while a Gemini answer in Spanish gave a range of 5 to more than 20. The cited Google documentation described query fan-out but, according to the article, did not supply either estimate. These figures should be understood as claims in the sampled answers—not verified descriptions of the systems’ internal behavior.
The article also reported a time-sensitive country-availability discrepancy: an AI Overview said the feature was available in more than 120 countries, while Google had announced availability in more than 200 countries and territories on May 20, 2025. Availability can change, so consult current official Google information before relying on either figure.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →What a citation does—and does not—prove
A link beside an AI answer is a lead to evidence, not proof that the evidence supports the claim. Open it and check the relevant passage. A source may be reputable yet irrelevant to the specific sentence, may support only part of it, or may have been superseded.
- Isolate the claim. Separate a precise number, date, recommendation, or attribution from the surrounding explanation.
- Open the cited source. Find the passage or data that directly supports the statement; do not rely on the link title or a search-result snippet.
- Check scope and date. Confirm that the source covers the right product, region, version, population, and time period.
- Look for stronger evidence if needed. For consequential claims, prefer the originating study, official documentation, or another primary source over a summary.
- Revise your confidence. If you cannot locate support, treat the claim as unverified rather than repeating it as fact.
OpenAI’s ChatGPT Search help page, as quoted by Tony G., advises: “Open a cited source to check that it supports the answer.” The practice applies beyond ChatGPT: a citation should be evaluated against the sentence it accompanies.
Recommendations and platform details need extra scrutiny
Advice can sound specific and authoritative without being grounded in current official guidance. Tony G.’s article reported that Gemini recommended FAQPage and HowTo markup and 40-to-60-word answer blocks. The article contrasted that with Google guidance saying no special schema.org markup was needed and reported that FAQ rich results stopped appearing in Google Search on May 7, 2026. Search features and guidance change; check current Google Search Central documentation before acting on such advice.
The test also reported source-credit and personalization observations: only Claude credited the 2023 paper that introduced GEO; none of ten answers about retrieval-augmented generation credited the 2020 paper that named it; two Spanish answers used names for Google AI Overviews that Google does not use; and 12 of 16 ChatGPT answers reflected account-related context despite temporary chats. These are observations from one sample, not universal product behavior.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesHow to use AI answers safely
- Verify high-impact specifics. Check important numbers, dates, eligibility rules, safety guidance, and recommendations against primary sources.
- Check freshness. Interfaces, country coverage, search features, and platform policies can change; confirm current official guidance when the answer depends on them.
- Compare languages when the decision spans markets. The test’s 32 answers per language offer a reason to check, not proof of a settled English-versus-Spanish difference.
- Benchmark cautiously. If comparing systems, use the same prompts and settings, document dates and context, and run prompts repeatedly. A single run can reveal issues worth investigating but cannot establish a dependable ranking.
There is a broader reason source quality matters. Tony G. also relayed statistics from other studies: a European Broadcasting Union/BBC review reportedly found serious sourcing problems in 31% of more than 3,000 news answers; an audit of four generative search engines reportedly found that 74.5% of citations supported the sentence they accompanied; and Pew Research Center findings were cited on AI summaries and clicks. Since those underlying studies were not independently checked for this article, they should be treated as reported context, not as independently verified measurements here. The same principle applies to any statistic in an AI answer: trace it to the original study before relying on it.
Quick Recap
Best Value
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.
