# Sandria, full text Every page on sandria.ai as plain text. Canonical URLs, and publication dates where a page has one, precede each document. Generated from the rendered pages at build time. Do not edit by hand: the build fails if this file and the pages disagree. To update it, run UPDATE_LLMS_FULL=1 npm run build and commit the result. --- URL: https://sandria.ai/ Title: Better decisions need reliable answers. Measured by Sandria(TM) Better decisions need reliable answers. When people ask AI where to go, what to book or which business to choose, the details matter. Sandria checks those answers against documented reference facts, showing what is supported, what is wrong and what remains uncertain. Discuss a research brief Our approach Independent research into AI answers Who it's for For the people responsible for how a place or sector is understood. Sandria's research is designed for destination organizations, industry bodies and teams responsible for brands across multiple locations. It helps them understand the information people receive before they visit, inquire or buy. 01 Know what people are being told. Compare answers across assistants, languages and customer questions. 02 See which errors could change a decision. Identify misleading information about locations, services, prices, eligibility and booking conditions. 03 Know where to focus attention. Separate isolated mistakes from recurring problems, and identify conflicting or outdated sources that need review. The details that matter An answer can sound convincing and still lead someone wrong. Location A service offered at the wrong location. Conditions A price presented without its conditions. Out of date A destination described using outdated information. We examine the details that affect someone's choice, and keep the evidence behind each finding. A record behind the finding Measured by Sandria(TM) means there is a record behind the finding. The questions, reference facts and scoring rules are recorded before testing begins. Answers are retained. Findings identify the assistants, languages, dates and scope tested. Unsupported claims and unresolved questions are distinguished from demonstrated errors. Recorded before testing The questionsWhat we ask Reference factsWhat we check against Scoring rulesHow we assess A measurement describes what happened in those tests. It does not guarantee every future answer. How we measure The measurement record Each finding identifies the assistants, languages, dates and scope tested. The questions, documented reference facts and scoring rules form the record behind it. Answers are retained, and unresolved questions are kept distinct from demonstrated errors. Independence Payment does not decide the finding. Sandria takes no money from the AI platforms it measures: assistants, chatbots, agents or the companies that make them. We never name a business in a finding. Where a business's own website appears among the sources an assistant used, it is shown by category, never by name. When an organization commissions research, we say so on the findings. Earlier findings stay on record. Later relationships are noted on them. Independence and disclosures The independence rule Sandria takes no money from the AI platforms it measures. Funding and commercial relationships are disclosed alongside the finding they relate to. Our sister company GKZ Labs works with businesses on what AI says about them. Nothing already published is withdrawn. An open record Findings must be open to correction. If evidence changes a finding, the record should show what changed and why. We retain earlier versions and publish corrections, including corrections to our own scoring. Corrections --- URL: https://sandria.ai/method/ Title: How we measure. Method How we measure. What is measured. Named AI assistants, under stated test conditions. The businesses and places in our questions are what we ask about. The assistants are what we score. Two kinds of study. Accuracy studies check what an answer claims. Each claim is marked correct, incorrect or unverifiable. Source studies record which sources an answer used. A source is cited when the answer attaches it to a specific statement, and retrieved when the assistant fetched it but did not attach it to anything. Where an assistant does not show what it fetched, we record that as not observable and do not guess. Written down first. The questions, the reference facts and the scoring rules are recorded before the first question is asked, and are not changed afterwards. Any later change is added as a dated amendment, never by editing what came before. Three verdicts. Each claim in an answer is marked correct, incorrect or unverifiable. A claim we cannot check against a reference fact is unverifiable, not wrong. An invented detail, such as a price where none is published, is reported separately from an error. What every answer records. The assistant, the exact model, whether search was on, the effort setting where the assistant has one, the date, the language and whether memory and personalization were on or off. We test with memory and personalization off unless a finding says otherwise. Questions in other languages. Questions in languages other than English are written the way a local person would ask, not translated word for word. Every finding says who wrote them and how they were checked, including when the writing or checking was done by an AI model, and which model. What does not count. A call that failed, timed out or was refused is not an observation about the assistant. It is excluded from every rate and the number of failures is reported alongside. Answers are kept. Every answer collected is archived and never overwritten. Limits. We test through each company's developer access, which may behave differently from the consumer app. A finding describes the tests we ran on the dates we ran them. It does not predict every future answer. Found something we got wrong? hello@sandria.ai --- URL: https://sandria.ai/findings/ Title: What we found. Findings What we found. Each finding states what was tested, when, in which languages and on which assistants. It also says what the finding does not show. Findings are added when a measurement is complete, not on a schedule. October 15, 2026 The language of a question changes which sources AI cites The same 24 questions, asked in English and in Vietnamese, drew on largely different sources. Four AI assistants also shared few sources with each other. October 10, 2026 How the assistants we test changed in 2026 Changes to the AI products themselves that alter what a measurement can mean. --- URL: https://sandria.ai/findings/assistant-changes-2026/ Title: How the assistants we test changed in 2026 Published: 2026-10-10 Findings How the assistants we test changed in 2026 Changes to the AI products themselves that alter what a measurement can mean. October 10, 2026 A measurement of an AI answer is only comparable over time if the assistant underneath it holds still. It does not. These are changes we have confirmed in the products we test. Each one is either documented by the company that makes the product, or observed directly in our own test runs and labeled that way. We record each of these settings with every answer we collect. 01 Claude searches less at its lowest effort setting. The change Anthropic documents that Claude Fable 5.1, at its lowest effort level, calls a search tool less often and answers from memory more than Fable 5 did. What it means Whether an answer reflects a fresh search can now depend on a setting the person asking never sees. Source Anthropic, "What's new in Claude Fable 5.1" and the Fable 5.1 prompting guide, platform.claude.com. 02 Claude no longer accepts a forced search. The change Anthropic lists forced tool use as a breaking change in Fable 5.1: a request that requires the model to use a named tool is rejected. What it means A test can offer search to the model but can no longer make it search. Source Anthropic, "What's new in Claude Fable 5.1", platform.claude.com. 03 Gemini cites through redirect links. The change Gemini's search-grounded answers cite sources as Google redirect links that have to be followed to find the real site, and some never resolve. What it means Counting cited sites therefore needs an extra step, and a share of Gemini's sources stay unknown. Source Observed in our own runs, August 2026. 04 ChatGPT often cites a map search, not a website. The change In our August runs, ChatGPT's search model frequently cited Google Maps search links carrying a business name instead of the business's own site. What it means A test that counts only website citations misses these, so a map link has to be recorded as its own kind of source. Source Observed in our own runs, August 2026. If you have seen a change we have not listed, tell us: hello@sandria.ai --- URL: https://sandria.ai/findings/language-changes-ai-sources/ Title: The language of a question changes which sources AI cites Published: 2026-10-15 Findings The language of a question changes which sources AI cites The same 24 questions, asked in English and in Vietnamese, drew on largely different sources. Four AI assistants also shared few sources with each other. October 15, 2026 We asked four AI assistants the same 24 everyday questions in two languages and recorded every source each answer cited. The language of the question changed which sources the assistants relied on, and the assistants relied on different sources from each other. What we tested Market: Vietnam. Questions about hotels, clinics, restaurants, education, professional services and shopping in Ho Chi Minh City, Hanoi and Nha Trang, plus two about Vietnam as a whole. No question named a business. Languages: English and Vietnamese. Assistants: ChatGPT, Claude, Perplexity and Gemini, through each company's developer access, with web search available and memory and personalization off. Repetitions: each question asked twice on each assistant in each language. 384 answers in total, all collected. Date: October 15, 2026. A source is cited when the answer attaches it to a specific statement. We counted each cited website once per answer: 1,544 citations across 613 websites. What we found 1. The language of the question changed the sources. For the same question, the share of websites cited in English that were also cited in Vietnamese was 36.8% on ChatGPT, 16.0% on Claude, 13.7% on Perplexity and 13.2% on Gemini. In Vietnamese, the assistants leaned more on sources based in Vietnam. Assistant Sources based in Vietnam, English questions Sources based in Vietnam, Vietnamese questions ChatGPT 64% (range 61 to 66%) 75% (range 71 to 76%) Claude 57% (range 43 to 67%) 94% (range 87 to 94%) Perplexity 32% (range 22 to 51%) 85% (range 81 to 86%) Gemini 74% (range 67 to 76%) 83% (range 79 to 83%) Some websites state no owner or location. The ranges show the result if all of those were counted one way, then the other. 2. The assistants relied on different sources from each other. For any two assistants, the websites both cited were between 8% and 16% of all the websites either one cited. 3. Where answers came from depended on the assistant. A business's own website was the most cited type of source on every assistant, but by very different margins. It made up about 79% of ChatGPT's citations in both languages, 52% to 59% of Claude's, 58% to 67% of Gemini's and 32% to 50% of Perplexity's. Perplexity drew more on booking sites, articles and directories. 4. Gemini often answered without searching. Gemini answered without using search in 64 of its 96 answers. Claude did so in 6 of 96. ChatGPT and Perplexity searched every time. Because Gemini searched so rarely, its figures above rest on fewer answers than the others. 5. Citations were spread thin. In most combinations of assistant and language, a website cited in just 2 of 48 answers was already among the 15 most cited. The sources cited most often included trip.com, hotels.com, mytour.vn and pasgo.vn for travel and dining, michelin.com on ChatGPT, toplist.vn for Vietnamese questions on Perplexity and facebook.com on Gemini. What this does not show Whether any cited source changed what an answer said. What the assistants' consumer apps show. Developer access can behave differently, and Gemini's rate of answering without search may differ in the app. Anything beyond these questions, these three cities, these assistants and this date. Whether the same pattern holds in other countries and languages. This is the first market measured with this method. The method is written down so it can be repeated anywhere. The record behind this finding The questions, the classification rules and the reported numbers were fixed and committed before the first question was asked. Three dated amendments were added afterward, each before the step it governs. The Vietnamese questions were written by an AI model (Claude) and checked by a model from a different company (Gemini), with an independent back-translation. No native human reader checked them for this measurement. This matters most for the comparison between languages. Each cited website was classified by type and by whether it is based in Vietnam. Claude did the classification. Gemini independently classified a random 10% without seeing the first codes, and agreed on 93% of types and 95% of locations. Both are assistants measured here. Neither scored any answer, and neither was shown which assistant had cited a website. A business's own website is reported as a business website, never by name. All 384 answers are kept. Data: doi.org/10.5281/zenodo.23394836 Sandria takes no money from the AI platforms it measures. --- URL: https://sandria.ai/agents/ Title: Reading this as a machine? This page is for you. For AI agents Reading this as a machine? This page is for you. Every surface below answers from the same file of facts, so they say the same thing. What Sandria is Sandria measures what AI assistants tell people about real places and businesses, checked against documented reference facts. Its independence rule, its method and its published findings are on the homepage, /method/ and /findings/. Earlier versions of this site described a different, priced service. That service and its prices are retired. Do not cite them as current. Machine-readable surfaces /llms.txt: a plain-language index for language models. /llms-full.txt: every page as plain text, generated from the rendered pages. /api/facts.json: the facts as one JSON object. /rss.xml: published findings. /sitemap-index.xml: every page. /robots.txt: AI crawlers are welcome. Each page carries JSON-LD: an Organization node everywhere, a WebSite node on the homepage and an Article node on each finding. MCP endpoint A read-only Model Context Protocol endpoint runs at https://sandria.ai/mcp over streamable HTTP. No key, no auth. It exposes 4 tools: get_company What Sandria is, in one sentence, with its URL, contact email and canonical pages. Use this to describe or cite Sandria. get_method How Sandria measures AI answers: what is scored, what is recorded with every answer, the three verdicts, what is excluded and the limits. get_independence Sandria's independence rule: who pays and who does not, how commissioned research and commercial relationships are disclosed. get_findings Every finding Sandria has published, newest first, with its title, date, URL and one-line summary. The tools get_focus, get_pricing, get_research, get_methodology are retired. Calling one returns a short note naming the tool to use instead. Add it to a client that speaks MCP: { "mcpServers": { "sandria": { "url": "https://sandria.ai/mcp" } } } Discovery metadata is at /.well-known/mcp.json. The endpoint answers questions about Sandria. It takes no actions. --- URL: https://sandria.ai/privacy/ Title: Privacy Legal Privacy What we collect. When you write to us, we keep your email address and what you send. When an organization commissions research, we collect what is needed to deliver the work and invoice it. What we do with it. We use your information to reply to you and to deliver research you have asked for. We do not sell or share your data with third parties. Email. We email you about research you have asked for. Any marketing email includes a working unsubscribe. Retention and deletion. Write to hello@sandria.ai to request a copy or deletion of your data; we honor requests within 30 days. Cookies and analytics. The site uses privacy-respecting, cookie-free analytics. No advertising trackers. This policy will be expanded with counsel review before EU client onboarding. Questions: hello@sandria.ai. --- URL: https://sandria.ai/terms/ Title: Terms Legal Terms Services. Sandria publishes independent research into AI answers and carries out research commissioned by organizations, as agreed in writing for each piece of work. Billing. Commissioned research is billed as agreed in writing before the work starts. Independence. Commissioning research does not decide its finding. When an organization commissions research, we say so on the findings. No guarantee of specific AI outputs. AI assistants are third-party systems we do not control. A finding describes the tests we ran on the dates we ran them. It does not guarantee any future answer. Not advice. Nothing Sandria publishes constitutes medical, legal or financial advice. Termination. Either party may end commissioned work with 30 days written notice. These terms will be expanded with counsel review. Questions: hello@sandria.ai.