Los Angeles restaurant websites and online ordering, 2026

What the homepages have

Each bar below is the share of the 376 homepages on which the study found that measure.

Almost every homepage carries a mobile viewport tag (99 percent), which a phone needs to show a page at its own width. A tap-to-call link, which places a call with one tap, is on 34 percent of them.

A link to Instagram is on 69 percent of them, more than link to any of the ordering or reservation tools the study counted. Most link to or load no ordering tool the study knows: 68 percent of them. LocalBusiness structured data is on 30 percent: markup a guest never sees, through which a page can give search engines facts such as its address and opening hours.

Where guests can order and book online

A homepage can send an online order to an ordering tool that passes it straight to the restaurant, or to a delivery app, a marketplace that lists many restaurants and takes the order itself. An ordering tool of the restaurant's own is linked or loaded on 32 percent of the homepages, Toast most often, on 53 of them. A delivery app is linked on 13 percent, and on 6 percent it is the only online ordering the study saw.

DoorDash Storefront, the ordering page DoorDash gives a restaurant to link from its own site, counts here as the restaurant's own ordering, apart from the DoorDash marketplace, so the name shows up in both charts below. A homepage with more than one tool or app counts once in its share and once under each name in the chart, so a chart's bars can add up to more than the share.

For a table, 16 percent of the homepages link to or load an online reservation tool, OpenTable most often, on 25 of them. A phone number or an email address for bookings does not count.

How heavy the homepages are

A homepage's HTML is the document alone, without the files it loads. The median homepage sent 145 kB of it, and the chart below shows how the rest spread across the size bands.

HTML size is not the same as speed: a light document can still load heavy images and scripts once it arrives.

Phone speed was not measured in this run. Google's PageSpeed Insights turned the requests away at its daily limit for use without an API key. The Chrome UX Report answers only with an API key, and the run had none. So this page states no speed figure, and the sample's speed columns are empty.

What it means for a restaurant owner

Each measure here is a question to ask of your own homepage, such as whether a guest on a phone can order from it and whether its menu is a page or a file.

Where the only online ordering is a delivery app, each online order goes through a marketplace that lists many restaurants and takes the order itself, so the guest orders inside the app rather than from the restaurant. An ordering tool of the restaurant's own sends the order straight to the restaurant instead.

Not every restaurant or cafe takes reservations, so the share with an online reservation tool is no target to match.

A menu as a PDF opens as a separate file rather than a page of the site, and the study did not test how either reads on a phone, so the plainest test is to open your own menu on one.

None of this says how good a restaurant is or how its guests like to order: it says what its homepage offers a guest who lands there.

How the sample was made

The sample was drawn from the restaurants and cafes that OpenStreetMap, a map its users build and edit, places inside the boundary of the City of Los Angeles, as its data stood on October 21, 2026. That boundary leaves out the separate cities around and inside it, such as Santa Monica and Beverly Hills.

The map query returned 1,391 of them, and 1,387 list a website. Left out were 20 whose listed page is on a social network, a review site, a delivery app or another shared host rather than a site of their own, and 276 more listings of a website already kept, so the locations of a chain that share a website count once. That left 1,091 sites. A shuffle with a fixed seed put them in an order the same map data always repeats, and the first 500 were drawn.

A program that gives its own name and a contact page with every request asked for each sampled homepage, and only for a site whose robots.txt rules allowed it. Its requests began at least 1 second apart, followed no more than 5 redirects, waited no more than 10 seconds for an answer and fetched nothing past the homepage.

Of the 500 homepages, 430 answered. The rest were never read: 11 were refused by their site's robots.txt rules, 8 had a robots.txt that answered with a server error or a too-many-requests reply, 3 gave no answer in time and 48 were out of reach or failed another way. Of those that answered, 48 answered with an HTTP error. Another 2 ended on a shared platform's page, and 4 on a domain for sale or a placeholder page.

That left 376 homepages measured: 294 of restaurants and 82 of cafes, from 402 restaurants and 98 cafes drawn. Every measure but the secure address comes from each homepage's HTML as its server sent it, before any script could change it; that one reads the address the homepage's redirects end on.

• Secure address (https): The homepage ends on an https address once the redirects from the root of the address the map lists are followed.
• Mobile viewport tag: The page has a viewport meta tag that sizes it to the screen (width=device-width or an initial scale), which a phone needs to show the page at its own width.
• Tap-to-call link: The homepage has a tel: link with a phone number, so a visitor on a phone can call with one tap.
• Own online ordering: The homepage links to or loads an online ordering tool that sends the order straight to the restaurant, such as Toast or ChowNow, named by its web address or script.
• Delivery app link: The homepage links to a delivery app: a marketplace that lists many restaurants and takes the order itself, such as DoorDash or Uber Eats, or a catering marketplace. A link anywhere on the page counts, the footer included.
• Orders only through delivery apps: The homepage links to a delivery app, and the study saw no ordering tool of the restaurant's own on it, so the only online ordering the study saw goes through a marketplace.
• Online reservation tool: The homepage links to or loads a known tool for booking a table or joining a waitlist online, such as OpenTable or Resy, named by its web address or script; a phone number or an email address for bookings does not count.
• Menu as a PDF: The homepage links to or embeds a PDF file that is a menu: the file's name or the link's words say menu, or name a meal or drinks, such as dinner or wine list, and no other kind of document, such as a job application. A menu written as a page of the site or posted as an image does not count.
• LocalBusiness structured data: The page carries JSON-LD structured data whose type is LocalBusiness or one of its schema.org subtypes, such as MedicalBusiness or Restaurant.
• FAQ structured data: The page carries JSON-LD structured data of the type FAQPage, which marks up questions and their answers.
• Chat widget: The page loads a known live chat or chatbot widget, named by its script.
• Mentions AI chat: The page's words mention AI chat or a chatbot, or the page loads a chat tool that runs on an AI model.
• Instagram link: The homepage links to an Instagram profile or post.
• HTML size: The size of the homepage's HTML document alone as fetched, without the files it loads, in kilobytes of 1,024 bytes.

The study's limits:

• Only restaurants and cafes that OpenStreetMap shows with their own website could be drawn, which is not every restaurant and cafe in the City of Los Angeles: one the map leaves out, or shows with no website, is not in the sample, and the map is more complete in some neighbourhoods than in others.
• Every share describes the 376 homepages measured, and another draw from the same map would give slightly different numbers: read any share as uncertain by up to about 5 percentage points.
• A tool is counted whenever its own link or script tag is in the HTML the server sends, even if its button or window only appears later; a tool that a different script brings in, a tag manager for example, never shows in that HTML and can be missed, so the counts for such tools are a floor.
• The study counts a vendor only from its own list, and a homepage whose tool is missing from that list counts as having none.
• Some website builders put online ordering on the restaurant's own address without naming an ordering vendor, which the study cannot see, so that ordering counts as none found.
• The share whose only online ordering the study saw is a delivery app can be off either way: own ordering the study cannot see makes the true share lower, and a delivery link that a script draws in later makes it higher.
• A menu counts as a PDF only when the homepage itself links to or embeds the file; a PDF linked from a separate menu page was not read.
• Places the map tags as fast food carry a separate tag and are not in this study.
• A domain for sale whose page says so only in its title or its scripts can still pass for a homepage.

The data

Every yes-or-no and vendor figure on this page can be counted again from the sample below; it rounds each HTML size, so the size median and bands come from the study's unrounded summary, published at the sample's address with summary.json in place of sample.csv.