Your Hero Video Starts Spending the Guest's Data Before They Read a Dish Name

The guest wanted a table for four at eight, the address, and whether the kitchen does anything a vegetarian would actually want to eat. What they are handed first is a download nobody requested, on a connection they pay for themselves. On the office broadband where the site was signed off, the hero looks like confidence. On a phone at a bus stop it is the restaurant helping itself to something belonging to the person it is trying to win.

The expensive part is that none of this ever comes back. A guest who waits, gives up and taps back leaves no enquiry, no message and no complaint. They go to the next listing, very often a directory or a marketplace that loads quickly because it has everything to gain from the click. The restaurant sees an ordinary quiet Thursday and no reason to think anything happened. Somebody else paid for the opening film, and the booking went to whoever charges commission on it.

The mechanism is not subtle, and it is written down in the reference documentation for the element that carries the film. A video on a web page has a preload attribute: the page's hint about how much of the file to fetch before anybody presses play. The specification advises browsers to set it to metadata, meaning the length and the dimensions rather than the film itself. Autoplay does not negotiate with that hint. It overrules it:

Read that as an owner rather than as a developer. One word in the markup outranks the page's own caution about not fetching a film before anybody has asked for it. Somebody — an agency, a template, a tick-box labelled "play automatically" — told the page this film is what the guest came for, and the browser is doing as instructed.

It is worth knowing that browsers have at times been rather protective about video on a mobile connection. Google's media-preloading guidance, last updated in August 2017 and now carrying an editorial warning from its own publisher that browser support has changed since publication, hedged its list as the rules applying at the time of writing, and one of them was about the connection the guest happens to be on:

That is a browser deciding, unprompted, that somebody on mobile data should not have a video fetched at them speculatively — in a snapshot now nine years old, from the same page that dates the metadata default to Chrome 64. Whether a browser shipping today still applies that cellular rule, and how it reconciles the rule with an autoplay attribute pointing the other way, was not established in this research. What the documentation does establish is narrower, and quite enough for an owner: autoplay takes precedence over the page's own preload hint, so the caution written into the markup is overruled by the markup. Even that is not absolute — the same MDN page records that a video carrying a lazy loading attribute will not start downloading and autoplaying until the element is near or within the viewport. Autoplay is an instruction, switched on by whoever built the page rather than by anybody thinking about the guest holding the phone.

Restaurant owners tend to imagine the guest's connection as a version of their own, and the office connection has no meter on it. Ofcom's pricing and consumer engagement work, published on 26 February 2026 and last updated on 30 June 2026, measured what pay-monthly mobile customers in this country actually consume:

And, on how much of what they pay for they use:

Ofcom published those figures to make a hedged consumer point: that many customers may be able to save money by choosing a tariff with an allowance that better matches what they need. Seen from a restaurant's side of the counter, it is the first figure that bites. Nearly three in ten pay-monthly users are living inside 2 GB a month — maps, messages, the work email that will not stop, and the twenty seconds they are about to spend deciding where to eat. The 55% figure describes people with headroom today; whether any of them move to a smaller tariff, and what that would leave for a hero video, Ofcom did not measure, and the step from that figure to a squeezed allowance belongs to this article rather than to Ofcom.

None of those guests will ever audit a restaurant's page. They do not know what a hero video weighs and they will never look. What they experience is a phone that goes warm, a page that will not settle, a first screen showing a dark rectangle while something loads behind it, and a vague sense that this place is hard work. On a tariff that has already flashed a warning once this month, the same twenty seconds reads as something closer to rudeness.

Honesty about the size of the hole matters more here than a confident number would. How many megabytes a typical independent restaurant's hero loop actually spends before a guest has tapped anything was not located in this research, and no figure for it is offered here — what follows rests on the documented behaviour of the autoplay attribute rather than on a measurement of any particular site. A byte count across live restaurant sites over a throttled mobile profile was planned and not carried out.

What does not depend on that measurement is the shape of the thing. Autoplay converts a file the guest might have chosen into a file the guest is given. A loop converts a single cost into a recurring one for as long as the page is open. And a moving file is larger than a still of the same scene — reasoning any owner can check on their own page, and reason enough to decide about the hero deliberately rather than by default.

Not a project. A check, done once, by whoever has the phone:

Two further checks belong to the same phone and the same thumb rather than to this article. The first is how many steps stand between the order button and allergen information — the subject of a separate piece that counts the clicks on an order page. The second is the other half of the pictures-instead-of-words problem: what happens when the menu itself is a photograph rather than structured content, which Google's own page about reading a menu off a photograph describes in terms an owner should read before relying on it.

The strongest inference in this article is that an autoplaying hero loop is spending a meaningful share of a real guest's data allowance. Autoplay establishes that the download starts without the guest's involvement, and Ofcom how small many UK allowances are; the bridge between the two — how many megabytes any particular restaurant's hero actually costs — was not located in this research, and the guest-side share therefore remains an inference rather than a measurement. How a browser shipping today reconciles an autoplay attribute with its own cellular preload policy was not established either: the preload rules quoted here come from a 2017 snapshot, hedged by its authors and since flagged by its publisher as overtaken. No prevalence figure is offered for how many independent restaurant sites autoplay video, because none was located. Search-demand evidence for the exact phrase this article targets is thin: an autocomplete observation supports the shorter prefix only.

What survives all of that costs nothing to act on. Open your own site on your own phone, with the Wi-Fi off, and watch what it takes from you before it tells you anything.