How AI reads this page
ChatGPT, Claude, Gemini and Perplexity don't look at colours, images or animations. They receive text, headings, links and structured data. Below is exactly what they read when they open “AI answers change every week: why your brand shows up and then disappears”.
The AI receives 1,911 words in 8 sections, 21 links and 3 blocks of structured data.
Text
The page content in Markdown, without menus or graphics. It is the format closest to how a language model reads text.
# AI answers change every week: why your brand shows up and then disappears > Three 2026 studies: the first name changes in 56% of weeks, two identical answers less than 1% of the time. What it means for measuring AI visibility. URL: https://geosnap.ai/en/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026 Language: English Version IT: https://geosnap.ai/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026 Author: Rinald Sefa, CMO Geosnap · Category: Research and data Publisher: Geosnap (Maind Group S.r.l.), https://geosnap.ai ## In short In 2026 three studies measured the volatility of AI answers: the brand named first changes from one week to the next in 56.4% of cases (Gemini 61.2%, ChatGPT 58.9%, Claude 49.1%), two answers to the same prompt return the same list less than 1% of the time, and platforms disagree on 27.2% of questions. A single answer is not data: what counts is the share of answers in which you appear, measured repeatedly, per engine and over time. - Across 480 question-engine pairs tracked for about three months, the first name changed in 56.4% of weeks and only 85 pairs kept the same leader for the whole period. - Repeating the same prompt 60-100 times, ChatGPT and AI Overviews return the same list of brands less than 1% of the time; yet strong brands appear in 60-90% of answers. - ChatGPT, Claude, Gemini and Perplexity disagree on whether to name a brand in 27.2% of 749 identical questions: visibility has to be measured engine by engine. - User history shifts recommended brands by 33.3 points in Gemini and 16.0 in ChatGPT; APIs recover only 79% of the brands seen in real sessions. - Measure share of answers with repetitions and 4-8 week time series, several phrasings per intent, one figure per engine. Written by the Geosnap AI agent, reviewed and approved by Rinald Sefa. Translated from the Italian original. [Read the original](https://geosnap.ai/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026) Ask ChatGPT "what is the best management software for a real estate agency", read the answer, note who shows up. A week later ask the same question and the list is different. It is not an impression: three studies published between January and October 2026 measured how much AI engines' recommendations change over time, across platforms and between users. Here are the numbers, the causes and what changes in how you measure your brand's visibility. ## How much do AI answers change from one week to the next? More than half the time, the "first name" changes. A New York agency asked [160 buying questions to ChatGPT, Claude and Gemini every week from April 27 to August 3, 2026](https://ecommercefastlane.com/ai-assistants-change-top-pick-weekly/), keeping one fixed model version per assistant: 480 question-engine pairs, 6,385 usable answers, 1,664 companies named. The result: the company named first changed from one week to the next in [56.4% of cases](https://ecommercefastlane.com/ai-assistants-change-top-pick-weekly/), and 82.3% of the pairs saw their top answer change at least once. Only 85 of the 480 pairs kept the same leader for the whole period. Volatility is not the same everywhere. In the same research [Gemini changed its leader in 61.2% of weeks, ChatGPT in 58.9%, Claude in 49.1%](https://ecommercefastlane.com/ai-assistants-change-top-pick-weekly/). One limit the authors state: the study did not log the sources retrieved for each answer, so it cannot tell a change caused by different pages from the randomness of text generation. | AI engine | Weeks in which the first name changes | | --- | --- | | Gemini | 61.2% | | ChatGPT | 58.9% | | Claude | 49.1% | | Average of the three | 56.4% | ## Are two answers to the same question identical? Almost never, not even on the same day. In January 2026 SparkToro and Gumshoe had hundreds of volunteers run [2,961 prompts on ChatGPT, Claude and Google Search's AI](https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242), including 12 brand or product recommendation requests repeated 60-100 times per platform. ChatGPT and AI Overviews returned the same list of brands [less than 1% of the time](https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242); the same list in the same order, less than 0.1%. Claude was slightly more stable on names, less so on order. The study was not peer-reviewed and one of the two authors sells tracking tools, which the authors disclose themselves. The most useful finding is a different one, though. Lists change, but strong names persist: for a given intent, the leading brands appeared in [60-90% of responses](https://searchengineland.com/ai-recommendation-lists-rarely-repeat-study-468076), whether the topic was hospitals, agencies or consumer products. In the 994 responses to headphone prompts written freely by 142 participants, the four main brands appeared in [55-77% of cases](https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242). So a single answer is a roll of the dice, but across many answers a stable frequency emerges. That frequency is the number to watch. ## Do the platforms agree with each other? In roughly three cases out of four yes, in the rest no. An analysis of [749 identical product-selection questions](https://www.digitaljournal.com/pr/news/access-newswire/new-research-llm-scout-reveals-1248216614.html), extracted from nearly 27,000 answers collected over nine months, found that ChatGPT, Claude, Gemini and Perplexity disagreed on whether to name a tracked brand in [27.2% of cases](https://www.digitaljournal.com/pr/news/access-newswire/new-research-llm-scout-reveals-1248216614.html). ChatGPT was the engine most likely to name a brand when the others stayed generic. It is a press release from the company that sells the tool used to collect the data, to be read with that limit in mind, but the picture matches the differences between engines seen above. For you this means "we are visible in AI" is not a meaningful sentence without saying in which engine. An office furniture manufacturer can show up regularly in ChatGPT and almost never in Gemini, and the two have different causes: different sources, different weight given to reviews, company sites, forums. It pays to measure each engine separately; Geosnap does this for ChatGPT, Gemini and Google AI Mode, with Perplexity and Claude coming next. ## Why does the same question give different answers? For three reasons that add up: text generation is probabilistic, the retrieved sources change, and the engine knows different things about each user. The first explains why two answers in the same minute differ. The second explains week-to-week changes: a new page, a review, a trade article enters the index and shifts the answer. The third is the least measured and the most interesting. A September 2026 study compared [8,609 responses across four platforms in the UK](https://www.einnews.com/pr_news/944542004/new-study-finds-ai-personalization-can-change-which-brands-chatgpt-and-gemini-recommend), with six user personas, three consumer categories and 30 prompts, under three conditions: history-free APIs, temporary logged-in sessions and accounts "primed" with a conversation history. Net of normal variation, the user's history shifted the set of recommended brands by [33.3 percentage points in Gemini, 16.0 in ChatGPT, 7.2 in Perplexity](https://www.einnews.com/pr_news/944542004/new-study-finds-ai-personalization-can-change-which-brands-chatgpt-and-gemini-recommend); in Claude no detectable effect. Accounts with a British history saw UK/EU-origin brands grow by 6.6 points and citations from .uk domains by 6.0 points. The authors warn that this does not prove localising a site automatically improves visibility. One technical detail affects how you measure. In the 173 matched comparisons, API calls recovered [79% of the recurring brands](https://www.einnews.com/pr_news/944542004/new-study-finds-ai-personalization-can-change-which-brands-chatgpt-and-gemini-recommend) seen in history-primed sessions, against 95% for temporary logged-in sessions. In other words, measuring via API is not the same as measuring as a real user: that is why [the two methods give different results](https://geosnap.ai/en/blog/api-vs-browser-la-differenza-api-e-browser-nei-metodi-di-formulazione-domanda). ## How do you measure AI visibility if answers keep changing? With repetitions, frequencies and time series, not with a single check. Four practical rules. 1. **Do not trust a single answer.** With less than 1% identical lists, a screenshot says little. Repeat each prompt several times and count in how many answers you appear: the share of answers is the number that holds up over time. 2. **Measure the trend, not the day.** If the first name changes in 56% of weeks, a drop in a single reading is noise. Look at the share of answers over four to eight weeks before deciding something has really changed. 3. **Keep the engines apart.** A 27% disagreement between platforms means your brand can do well in one and badly in another. Keep one figure per engine and look for the causes in the sources each one cites. 4. **Choose prompts carefully.** In the test with 142 people, freely written prompts for the same need had a semantic similarity of just [0.081](https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242): people ask the same thing in very different ways. Monitor several phrasings per intent, not one. [Here is how to choose the questions to track](https://geosnap.ai/en/blog/quali-prompt-monitorare-risposte-ai-come-scegliere). On the content side, volatility is also an opportunity: if lists reshuffle every week, a brand that is absent today has more chances to get in than it would on a results page that stays still for months. The things that raise citation frequency are the usual ones: pages that answer directly, presence in the sources engines read, such as reviews, comparisons and trade publications, and up-to-date content. In Geosnap reports the share of answers per prompt and per engine is the number to follow week by week, together with the sources cited when the brand appears and when it does not. ## Sources - [Is ChatGPT Recommending Your Store? Why AI Assistants Change Their Top Pick Most Weeks](https://ecommercefastlane.com/ai-assistants-change-top-pick-weekly/) (eCommerce Fastlane (ricerca BusySeed), 2026-10-16) - [AI Recommendations Change With Nearly Every Query: SparkToro](https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242) (Search Engine Journal, 2026-01-30) - [AI recommendation lists repeat less than 1% of the time: Study](https://searchengineland.com/ai-recommendation-lists-rarely-repeat-study-468076) (Search Engine Land, 2026-01-30) - [New Research by LLM Scout Reveals ChatGPT, Claude, Gemini, and Perplexity Disagree on Brand Mentions in More Than One in Four Evaluation Queries](https://www.digitaljournal.com/pr/news/access-newswire/new-research-llm-scout-reveals-1248216614.html) (ACCESS Newswire (via Digital Journal), 2026-10-07) - [New Study Finds AI Personalization Can Change Which Brands ChatGPT and Gemini Recommend](https://www.einnews.com/pr_news/944542004/new-study-finds-ai-personalization-can-change-which-brands-chatgpt-and-gemini-recommend) (EIN Presswire (Cassie Clark Marketing, friction AI), 2026-09-23) ## Frequently asked questions ### If my brand does not appear in ChatGPT today, do I have a problem? Not necessarily. Two answers to the same prompt give the same list less than 1% of the time, so a single reading says almost nothing. Repeat the prompt several times and look at how many answers include you: if the share is consistently low for several weeks, then there is something to fix. ### Why does my brand appear in ChatGPT but not in Gemini? Because engines draw on different sources and weigh them differently. A study of 749 identical questions found platforms disagreeing in 27.2% of cases, and BusySeed's research shows Gemini is also the most volatile week over week. You need one figure per engine and an analysis of the sources each one cites when it does not name you. ### How many times do I need to repeat a prompt to get reliable data? The studies give no definitive number and SparkToro lists it as an open question. In practice you need enough repetitions to turn randomness into a stable frequency: in the research cited, leading brands appear in 60-90% of answers even though the lists almost always change. Ten or so readings a week over several weeks beat a hundred in one day. ### Are API answers the same as what a user sees? No. In the personalisation study, API calls recovered 79% of the recurring brands seen in sessions with a history, against 95% for temporary logged-in sessions. The user's history shifted the set of brands by 33.3 points in Gemini and 16.0 in ChatGPT. That is why Geosnap simulates a real user in the browser instead of querying APIs. ### Is the volatility of AI answers a risk or an opportunity? Both. It is a risk if you make decisions on a single screenshot. It is an opportunity because lists reshuffle every week: a brand absent today has more chances to get in than on a results page that stays still for months, if it works on the sources engines read and on pages that answer directly.
Structured data
Schema.org data (JSON-LD) embedded in the page code. It tells the AI explicitly what the page is about and who published it.
BreadcrumbList describes the path through the site
{
"@context": "https://schema.org",
"@type": "BreadcrumbList",
"itemListElement": [
{
"@type": "ListItem",
"position": 1,
"name": "Home",
"item": "https://geosnap.ai/en/"
},
{
"@type": "ListItem",
"position": 2,
"name": "Blog",
"item": "https://geosnap.ai/en/blog/"
},
{
"@type": "ListItem",
"position": 3,
"name": "Research and data",
"item": "https://geosnap.ai/en/blog/#ricerca-e-dati"
},
{
"@type": "ListItem",
"position": 4,
"name": "AI answers change every week: why your brand shows up and then disappears",
"item": "https://geosnap.ai/en/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026"
}
]
}
BlogPosting describes the article: author, dates and sources
{
"@context": "https://schema.org",
"@type": "BlogPosting",
"@id": "https://geosnap.ai/en/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026#article",
"headline": "AI answers change every week: why your brand shows up and then disappears",
"description": "Three 2026 studies: the first name changes in 56% of weeks, two identical answers less than 1% of the time. What it means for measuring AI visibility.",
"inLanguage": "en",
"url": "https://geosnap.ai/en/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026",
"mainEntityOfPage": "https://geosnap.ai/en/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026",
"image": "https://geosnap.ai/img/blog/p2026-10-23-volatilita-risposte-ai-en.webp",
"articleSection": "Research and data",
"wordCount": 1249,
"author": {
"@type": "Person",
"name": "Rinald Sefa",
"jobTitle": "CMO",
"url": "https://geosnap.ai/en/about",
"worksFor": {
"@id": "https://geosnap.ai/#organization"
}
},
"publisher": {
"@id": "https://geosnap.ai/#organization"
},
"isAccessibleForFree": true,
"abstract": "In 2026 three studies measured the volatility of AI answers: the brand named first changes from one week to the next in 56.4% of cases (Gemini 61.2%, ChatGPT 58.9%, Claude 49.1%), two answers to the same prompt return the same list less than 1% of the time, and platforms disagree on 27.2% of questions. A single answer is not data: what counts is the share of answers in which you appear, measured repeatedly, per engine and over time.",
"datePublished": "2026-10-23",
"dateModified": "2026-10-23",
"citation": [
"https://ecommercefastlane.com/ai-assistants-change-top-pick-weekly/",
"https://www.searchenginejournal.com/ai-recommendations-change-with-nearly-every-query-sparktoro/566242",
"https://searchengineland.com/ai-recommendation-lists-rarely-repeat-study-468076",
"https://www.digitaljournal.com/pr/news/access-newswire/new-research-llm-scout-reveals-1248216614.html",
"https://www.einnews.com/pr_news/944542004/new-study-finds-ai-personalization-can-change-which-brands-chatgpt-and-gemini-recommend"
],
"translationOfWork": {
"@id": "https://geosnap.ai/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026#article"
}
}
FAQPage describes the frequently asked questions
{
"@context": "https://schema.org",
"@type": "FAQPage",
"@id": "https://geosnap.ai/en/blog/risposte-ai-cambiano-ogni-settimana-brand-volatilita-dati-2026#faq",
"mainEntity": [
{
"@type": "Question",
"name": "If my brand does not appear in ChatGPT today, do I have a problem?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Not necessarily. Two answers to the same prompt give the same list less than 1% of the time, so a single reading says almost nothing. Repeat the prompt several times and look at how many answers include you: if the share is consistently low for several weeks, then there is something to fix."
}
},
{
"@type": "Question",
"name": "Why does my brand appear in ChatGPT but not in Gemini?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Because engines draw on different sources and weigh them differently. A study of 749 identical questions found platforms disagreeing in 27.2% of cases, and BusySeed's research shows Gemini is also the most volatile week over week. You need one figure per engine and an analysis of the sources each one cites when it does not name you."
}
},
{
"@type": "Question",
"name": "How many times do I need to repeat a prompt to get reliable data?",
"acceptedAnswer": {
"@type": "Answer",
"text": "The studies give no definitive number and SparkToro lists it as an open question. In practice you need enough repetitions to turn randomness into a stable frequency: in the research cited, leading brands appear in 60-90% of answers even though the lists almost always change. Ten or so readings a week over several weeks beat a hundred in one day."
}
},
{
"@type": "Question",
"name": "Are API answers the same as what a user sees?",
"acceptedAnswer": {
"@type": "Answer",
"text": "No. In the personalisation study, API calls recovered 79% of the recurring brands seen in sessions with a history, against 95% for temporary logged-in sessions. The user's history shifted the set of brands by 33.3 points in Gemini and 16.0 in ChatGPT. That is why Geosnap simulates a real user in the browser instead of querying APIs."
}
},
{
"@type": "Question",
"name": "Is the volatility of AI answers a risk or an opportunity?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Both. It is a risk if you make decisions on a single screenshot. It is an opportunity because lists reshuffle every week: a brand absent today has more chances to get in than on a results page that stays still for months, if it works on the sources engines read and on pages that answer directly."
}
}
]
}
Metadata
The information in the page header, which crawlers and engines read before the content.