Frequently asked questions

Something missing? Write to info@audioguideeverything.com.

Does it work offline?

No. Each guide is written on the spot, for that photo or that place, so the app needs an internet connection: mobile data or Wi-Fi.

Travelling abroad? A local SIM card or an eSIM is usually the cheapest way to stay connected.

Where does it work?

Anywhere in the world.

Explain a photo works wherever you are. Guide me here starts from what OpenStreetMap and Wikipedia say about the place, so its guides are richest where those are detailed, as in most towns and well-known sites. Where little is mapped or written about, the guide has less to tell.

How accurate are the guides?

Usually right, sometimes wrong: think of the app as a curious companion, not a reference book.

Guide me here starts from OpenStreetMap and Wikipedia, not from the AI's memory alone. But an AI model writes every guide, and it can get a detail wrong or misread a photo. Check before relying on it for anything that matters, like opening hours, safety or what's in a dish if you have an allergy.

If a guide is wrong, rate it and tell us what was off. You can also tick the box to send us a copy of the guide, so we can find and fix the problem.

Why are guides to photos from my library less precise?

Because the app often doesn't know where the photo was taken, and knowing the place helps it tell one statue or dish from another.

A photo taken in the app comes with your location. A photo from your library usually doesn't: Android removes it to protect your privacy. Android will soon let you share it, and we'll use that.

For the most precise guide, take the photo with the app, on the spot.

What is the environmental impact of the app?

One guide emits between 0.2 and 4 g of CO₂, as much as driving a car for 2 to 45 metres, and uses at most two-thirds of a glass of water. Narration, the costliest part, is made only when you ask for it, and only once per guide.

For one guide

Per guideLow estimateHigh estimate
Greenhouse gases~0.2 g CO₂e2 m by car15 m by train10 s of video streaming~4 g CO₂e45 m by car300 m by train5 min of video streaming
Electricity~0.5 Wh30 Wikipedia pages2.5% of a phone chargeNot published
Water~0.5 mLa few drops~170 mLtwo-thirds of a glass
Metals and mineralsNot published~0.6 mg antimony eq.

Over a two-week trip

Say you use the app a lot on holiday: 10 guides a day for two weeks, 140 in all. That adds up to:

  • 30 to 560 g of CO₂, as much as 300 m to 6 km by car, 2 to 45 km by train, or half an hour to 10 hours of video streaming
  • 70 mL to 24 L of water, from a quarter of a glass to about a hundred glasses
  • About 70 Wh of electricity (low estimate), enough to charge a smartphone three and a half times
  • Up to 85 mg antimony eq. of metals and minerals

How we estimated it

AI providers don't publish exact figures per request, so we give a range. We start from our own usage: on average, a guide sends about 1,500 tokens (pieces of words) to the AI and gets about 750 back. We then apply two reference studies.

  • Low estimate: Google's own measurement of a typical AI text prompt, 0.24 Wh and 0.26 mL of water. We recalculated its emissions with the average US electricity grid, where the data centres are.
  • High estimate: Mistral AI's life-cycle assessment of a much larger model, 1.14 g CO₂e, 45 mL of water and 0.16 mg antimony eq. per 400-token answer. It was carried out with the consultancy Carbone 4 and the French ecological transition agency (ADEME), reviewed by two independent auditors, Resilio and Hubblo, and follows the international standards (ISO 14040/44 and the GHG Protocol). It counts more of the chain, from making the servers to the water used to produce the electricity.
  • Narration: no figures are published, so we go by size. A one-minute narration is about 1,900 tokens of audio, two and a half times the text of a guide, and costs about three times as much to make, so we count it as three text requests. One guide in three is narrated, which about doubles the figures for text alone.

We can't say where in this range the real figures fall: the low estimate is for a typical request, probably shorter than a guide, and the high one for a much larger model. Neither counts training the AI models, sending the guide to your phone or your phone's own energy use. We will update this answer as providers publish better data.

What the comparisons are based on

  • Car: the average new car sold in Europe in 2025, on official test values. Real-world driving emits somewhat more.
  • Video: the European average for streaming.
  • Wikipedia: its servers used 3.1 GWh in 2021, when people viewed 192 billion pages, about 0.016 Wh a page. As for the guide, the network and your phone are left out.
  • Train: about 13 g CO₂e per passenger-km, the average of a French high-speed train (2.9 g), the average Swiss train over its whole life cycle (7 g) and the average British train in 2025–26 (28 g). They don't count exactly the same things: take it as an order of magnitude.
  • Metals and minerals: the standard life-cycle indicator for their depletion, rare earths included, in antimony equivalent.
Sources and further reading

Data used above

Further reading

  • EcoLogits: an open-source, peer-reviewed method and free calculator for the footprint of AI requests (paper)
  • ML.ENERGY Leaderboard (University of Michigan): the measured energy use of open AI models

Critical views

Last updated 8 November 2026.