Original research · Updated
We Asked 5 AI Chatbots How Much Water You Need. Here’s How They Did.
We asked ChatGPT, Gemini, Claude, Perplexity and Google’s AI Overviews 15 common hydration questions, 315 answers in all, and checked every one against the National Academies, CDC, FDA, NIH and ACOG.
By Hydrate · Sources: National Academies, CDC, FDA, NIH, ACOGAll 315 answers (CSV)Cite thisPress contact
6 of 21
Answers on breastfeeding that matched the guidance, the weakest topic.
Hydrate study
96%
Of ChatGPT’s answers were fully correct, the highest of the five.
72 of 75
0
Answers that claimed 75% of Americans are chronically dehydrated.
of 315
6 of 15
Questions Gemini answered the same way all five times, the least consistent.
Hydrate study
What we found
- None of the 315 answers was flatly wrong, and no assistant repeated the claim that 75% of Americans are chronically dehydrated or that coffee dehydrates you.
- ChatGPT was the most accurate: 96% (72 of 75) of its answers were fully correct.
- Breastfeeding tripped them up most: only 6 of 21 answers matched the guidance. Most gave the right number of cups (about 13) but said it includes water from food; it doesn't, so taken literally that's roughly 2.5 cups a day too little.
- Gemini was the least consistent: asked the same question five times, it got the same verdict every time on only 6 of 15 questions.
- Only 31% of the sources the assistants cited were government or academic. Claude cited a source in 4 of 75 answers.
How accurate is AI hydration advice?
We asked each assistant 15 questions people really search, five times each (Google's AI Overviews once, since they don't change between searches), then scored every answer against an answer key built from the National Academies, CDC, FDA, NIH and ACOG. “Partly correct” mostly means the right number with the wrong label, such as calling the amount to drink the total including food.
Share of answers that were fully correct, by assistant
315 answers, collected 2026-10-16. Number of answers in brackets.
| Row | Fully correct | Partly correct | Incorrect |
|---|---|---|---|
| ChatGPT (75) | 96% (72 of 75) | 4% (3 of 75) | 0% (0 of 75) |
| Claude (75) | 85% (64 of 75) | 15% (11 of 75) | 0% (0 of 75) |
| Google AI Overviews (15) | 73% (11 of 15) | 27% (4 of 15) | 0% (0 of 15) |
| Perplexity (75) | 72% (54 of 75) | 28% (21 of 75) | 0% (0 of 75) |
| Gemini (75) | 68% (51 of 75) | 32% (24 of 75) | 0% (0 of 75) |
Women's health questions were harder
Four questions were about women: daily needs, pregnancy, breastfeeding, and women over 60. ChatGPT, Gemini, Claude and Perplexity scored lower on these than overall. Breastfeeding was the weakest: 6 of 21 answers were fully correct. The usual slip was the same one many people make: the guideline counts water from food, so the amount to drink is about a fifth lower. See it in cups.
Women's health questions: fully correct, by assistant
Questions on daily needs for women, pregnancy, breastfeeding, and a 65-year-old woman.
| Row | Fully correct | Partly correct | Incorrect |
|---|---|---|---|
| ChatGPT (20) | 85% (17 of 20) | 15% (3 of 20) | 0% (0 of 20) |
| Google AI Overviews (4) | 75% (3 of 4) | 25% (1 of 4) | 0% (0 of 4) |
| Perplexity (20) | 55% (11 of 20) | 45% (9 of 20) | 0% (0 of 20) |
| Claude (20) | 50% (10 of 20) | 50% (10 of 20) | 0% (0 of 20) |
| Gemini (20) | 45% (9 of 20) | 55% (11 of 20) | 0% (0 of 20) |
Question by question
Share of answers fully correct, by question and assistant
Each cell is five answers (one for Google AI Overviews). Hover a cell for the count.
| Question | ChatGPT | Gemini | Claude | Perplexity | Google AI Overviews |
|---|---|---|---|---|---|
| How much water should a woman drink a day? | 100% | 80% | 100% | 100% | 100% |
| How much water should a man drink a day? | 100% | 100% | 100% | 0% | 100% |
| How much water should I drink a day? | 100% | 60% | 100% | 0% | 100% |
| Do you really need 8 glasses of water a day? | 100% | 100% | 80% | 60% | 100% |
| Are most Americans chronically dehydrated? | 100% | 80% | 100% | 100% | 0% |
| Does coffee dehydrate you? | 100% | 100% | 100% | 100% | 100% |
| How much water should I drink while pregnant? | 80% | 60% | 20% | 100% | 100% |
| How much water should I drink while breastfeeding? | 60% | 20% | 0% | 20% | 100% |
| How much water should a 65-year-old woman drink a day? | 100% | 20% | 80% | 0% | 0% |
| Does the water in food count toward how much water I need? | 100% | 60% | 100% | 100% | 100% |
| Is thirst a good way to know when to drink water? | 100% | 80% | 100% | 100% | 0% |
| How much more water do I need when it's very hot or I'm exercising a lot? | 100% | 60% | 100% | 100% | 0% |
| What percentage of the human body is water? | 100% | 100% | 100% | 100% | 100% |
| Can drinking more water help prevent kidney stones? | 100% | 100% | 100% | 100% | 100% |
| How much caffeine is safe to have in a day? | 100% | 0% | 100% | 100% | 100% |
Did they repeat hydration myths?
Rarely. We checked every answer for five common myths. Mentioning a myth to debunk it, or calling “8 glasses” a loose rule of thumb, didn't count.
Answers that repeated a hydration myth as fact
Share of each assistant's answers that stated at least one of the five myths as fact or official advice.
- ChatGPT0% (0 of 75)
- Gemini8% (6 of 75)
- Claude0% (0 of 75)
- Perplexity0% (0 of 75)
- Google AI Overviews13% (2 of 15)
Which myths came up
Number of answers, out of 315, that stated each myth as fact.
- “8 glasses a day” as a rule3
- “Half your body weight in ounces”2
- “75% of Americans are dehydrated”0
- “Coffee dehydrates you”0
- “If you’re thirsty, you’re already dehydrated”5
Ask twice, get a different answer?
Asking the same question again can get a different answer. We counted how many questions each assistant got the same verdict on all five times.
Questions answered consistently across five tries
Out of 15 questions. Google's AI Overviews were captured once per question, so aren't included.
- Perplexity87% (13 of 15)
- ChatGPT87% (13 of 15)
- Claude80% (12 of 15)
- Gemini40% (6 of 15)
Where AI gets its hydration advice
Sources the assistants cited, by type
1130 citations across all answers. Government or academic: .gov, .edu, NIH, the National Academies and similar. Major health publishers: Mayo Clinic, Cleveland Clinic, Harvard Health, Healthline, WebMD and similar.
- Government or academic
- Major health publishers
- Everything else
Most-cited websites
Number of answers citing each site at least once.
- mayoclinic.org42
- health.harvard.edu36
- health.clevelandclinic.org36
- healthline.com34
- ncbi.nlm.nih.gov27
- nationalacademies.org25
- health.osu.edu25
- cdc.gov24
- nutritionsource.hsph.harvard.edu21
- webmd.com20
Answers that cited any source
Share of each assistant's answers citing at least one source (answers with a source / all answers). Some assistants only search the web when they decide to; Claude and Perplexity were run with web search allowed.
- ChatGPT59% · 44/75
- Gemini19% · 14/75
- Claude5% · 4/75
- Perplexity100% · 75/75
- Google AI Overviews100% · 15/15
Questions people ask
Can you trust ChatGPT for health advice?
- On basic hydration questions, mostly: in our test 96% of ChatGPT's answers were fully correct and none of the 315 answers across five assistants was flatly wrong. They slipped on details that matter, like breastfeeding, so check anything important with a doctor.
Which AI chatbot is most accurate for health questions?
- In our hydration test: ChatGPT 96%, Claude 85%, Perplexity 72%, Gemini 68% of answers fully correct. That's 15 questions on one topic, not health advice in general.
How accurate is Google’s AI Overview?
- 73% of Google's AI Overviews were fully correct in this study. A follow-up across 40 searches found the same rate, and that the Overviews cited Facebook more often than the CDC.
Methodology
What we asked. 15 hydration questions, worded the way people search them, chosen and frozen before collection along with the answer key and scoring rules. Each answer key comes from a primary source we checked: the National Academies' Adequate Intakes, ACOG, the FDA, NIDDK, USGS, the National Athletic Trainers' Association, and the 2002 review of the “8 x 8” rule.
How we collected it. On 2026-10-16, through DataForSEO: ChatGPT and Gemini as the public apps return them (US, English; models reported as gpt-6 and 3.5 Flash-Lite), Claude (claude-sonnet-5-5) and Perplexity (sonar-pro) through their APIs with web search allowed, and Google's AI Overviews from US desktop search. Each question was asked five times per assistant in a fresh session, with no system prompt; AI Overviews were captured once.
How we scored it. Two independent AI judges (Claude), working blind to which assistant wrote each answer, scored every answer as fully correct, partly correct, incorrect or no answer, and flagged five myths. They agreed on 97% of answers. A third review decided the 11 disagreements and re-checked a random 10% of agreements, overturning none. Because the judges are Claude models and Claude is one of the assistants scored, every answer and verdict is published in the CSV.
A correction we made. Our first answer key for pregnancy only accepted the National Academies' figures. Most assistants instead gave ACOG's “8 to 12 cups a day”, which is real guidance, so we added it and re-scored all 21 pregnancy answers. This raised the assistants' scores.
Limits. One day of answers, in English, from the US. Assistants change often, and a different wording or follow-up question can change an answer. Differences between assistants are only called differences when a two-proportion test gives p < 0.05. Read the full method and change log and the questions and answer key.
Cite this report
You're welcome to use any figure or chart on this page. Please credit Hydrate and link back to this page so readers can check the sources.
APA
Hydrate. (2026, October 16). We Asked 5 AI Chatbots How Much Water You Need. Here’s How They Did.. https://hydrateai.app/data/ai-hydration-advice
MLA
"We Asked 5 AI Chatbots How Much Water You Need. Here’s How They Did.." Hydrate, 16 Oct 2026, hydrateai.app/data/ai-hydration-advice.
Link (HTML)
<a href="https://hydrateai.app/data/ai-hydration-advice">We Asked 5 AI Chatbots How Much Water You Need. Here’s How They Did.</a> (Hydrate)
Sources
- National Academies of Sciences, Engineering, and Medicine (2005). Dietary Reference Intakes for Water, Potassium, Sodium, Chloride, and Sulfate.
- National Academies (2004). Report Sets Dietary Intake Levels for Water, Salt, and Potassium to Maintain Health and Reduce Chronic Disease Risk.
- American College of Obstetricians and Gynecologists (2026). How much water should I drink during pregnancy? (Ask ACOG).
- U.S. Food and Drug Administration (2024). Spilling the Beans: How Much Caffeine Is Too Much?.
- National Institute of Diabetes and Digestive and Kidney Diseases (2017). Eating, Diet, & Nutrition for Kidney Stones.
- U.S. Geological Survey (2019). The Water in You: Water and the Human Body.
- Journal of Athletic Training (National Athletic Trainers’ Association) (2017). National Athletic Trainers’ Association Position Statement: Fluid Replacement for the Physically Active.
- American Journal of Physiology (2002). "Drink at least eight glasses of water a day." Really? Is there scientific evidence for "8 x 8"?.
- Snopes (2001, updated). Do You Need Eight Glasses of Water Per Day? (fact check).
- CDC National Center for Health Statistics (2016). Daily Water Intake Among U.S. Men and Women, 2009–2012 (NCHS Data Brief No. 242).
This page summarizes public data for general information. It is not medical advice; talk to a clinician about your own fluid needs, especially if you are pregnant, have kidney or heart conditions, or take medications that affect fluid balance.
More Hydrate data
Original research
We Checked What Google Tells You About Drinking Water
Original research
The South Googles Dehydration. The West Googles Electrolytes.
Data hub
Hydration Statistics 2026: 33 Facts From Government Data
State ranking
The Hottest States in the US, Ranked by Heat & Hydration Risk
Data report
How Much Water Should a Woman Drink? Women's Hydration Statistics
Data report
What Americans Drink: Water, Coffee, Soda & More