The web app manifest on 27 homepages: 9 declare one, and two never say the site's name

FIELD TEST · 2026-08-29 · 27 homepages · 9 manifests fetched and parsed

Sample and method: the same 30-site panel in use since 2026-08-15, each homepage fetched once on 2026-08-29 at 13:27 (+0800) with a browser user agent and no JavaScript execution. Every <link rel="manifest">, every icon link and every theme-color meta was recorded, and each declared manifest URL was then requested and parsed as JSON.

Nine of 27 homepages declare a web app manifest, all nine resolve, and two of the nine never say what the site is called. Measured against the installability list Chromium documents, four of the nine qualify. The file is where a fourth copy of your brand name lives, and it is the copy nobody proofreads.

How we measured it

One request for each homepage, then one request for each manifest it pointed at. The inclusion rule was fixed before the run: final status 200, decoded body of at least 10,000 bytes, body containing <body. stackoverflow.com and medium.com returned 403 and www.reddit.com returned an 8,393-byte shell, leaving 27 sites.

The checks come from the specification, not from us. W3C defines the file as "a JSON document that contains startup parameters and application defaults for when a web application is launched", and defines name as "a string that represents the name of the web application as it is usually displayed to the user". MDN lists what Chromium-based browsers require: name or short_name, an icons array that "must contain a 192px and a 512px icon", start_url, display or display_override, and prefer_related_applications "must be false or not present".

Who ships a web app manifest at all

Nine sites, and the split is not about company size.

The nine are developer.mozilla.org, vercel.com, linear.app, figma.com, railway.com, www.netlify.com, github.com, substack.com and www.bbc.com. Every publisher on the panel — www.nytimes.com, www.wired.com, arstechnica.com, techcrunch.com, www.theverge.com — ships none, while www.bbc.com does.

Icons are a different story from manifests. There are 121 icon link elements across the 27 homepages and 22 sites carry at least one apple-touch-icon, so more than twice as many sites care about the icon as care about the file that would describe it.

Four of nine would install

Against MDN's documented Chromium list, five of the nine fail, and each fails for a different reason.

The github.com row is the one to read carefully, because it is not a mistake. Setting prefer_related_applications to true is how a site says it would rather you installed its native app, and GitHub's manifest also carries a related_applications list. It fails the installability check on purpose. The other four look like files nobody has opened since they were generated.

All nine, including the broken ones, ship both a 192px and a 512px icon. That is the one requirement the generators get right, and it is also the one you would notice if it were wrong.

Two manifests do not name the site

This is the row with something at stake for search, and it costs one line of JSON to fix.

The www.netlify.com manifest contains exactly one key, icons. The www.bbc.com manifest declares name and short_name and leaves both as empty strings, which is the harder version of the same problem: a linter sees the key and moves on. Both files are served with a 200 and neither browser nor crawler complains.

A manifest is a machine-readable statement of who you are. Leaving the name blank is answering the question with silence.

One more oddity sits in the same column. linear.app sets start_url to https://linear.app/login — an absolute URL to a sign-in page. It passes every documented check, and it means the installed entry point is the login screen rather than the product.

Even the content type disagrees

MDN says the response "should return Content-Type: application/manifest+json". Five of nine do.

The application/octet-stream one is www.netlify.com again — the same file that has only an icons array. A static host serving an unrecognised extension as a byte stream is the ordinary explanation, and it is also the ordinary reason nobody noticed the file was empty of everything else.

What this means for you

Three lines of work, in this order.

1. Open your own manifest in a browser tab. Not the generator that made it — the file your server actually returns. Two of nine sites here would have caught their problem in that one step.
2. Make name match the brand string you use everywhere else, exactly. It is one more place a machine reads who you are, and the case for keeping every copy identical is set out in entity SEO for AI answers.
3. Check start_url points at something you want opened. Four of nine omit it entirely and one points at a login page.

None of this is a ranking factor and we are not going to imply it is. It belongs to the same family as the icon set — small declarations that machines read and humans never look at — and the icon half of that family was measured in favicon SEO across 27 homepages. If the more urgent question is whether crawlers reach and quote your pages at all, that is what QueryWin works on.

What this run cannot tell you

Whether any of these sites is actually installable in a real browser. We compared each file against a documented list; we did not open Chrome and try, and browsers apply extra conditions we did not evaluate. We do not know whether a search engine or an answer engine reads the manifest at all — no documentation we found says one way or the other, so the brand-name argument above is about consistency, not about a measured effect. And we requested only homepage manifests: a site could declare a different file elsewhere.

Common questions

How did you measure this?

One fetch per homepage on 2026-08-29, parsing the delivered HTML for rel="manifest", icon links and theme-color, then one fetch for each of the nine manifest URLs, parsed as JSON and compared against MDN's documented Chromium requirements.

Does a web app manifest help SEO?

Nothing here shows that it does, and we found no documentation claiming it. The honest reason to fix yours is that it is one more machine-readable copy of your name, and it currently disagrees with the site on two of nine.

What is the difference between name and short_name?

W3C defines short_name as "a short version of the name of the web application", intended "where there is insufficient space to display the full name". One of the nine here omits it.

Do I need a manifest if I am not building an app?

Not for search. Eighteen of these 27 sites do without one, including every news publisher on the panel except the BBC.

Which Content-Type should it be served with?

application/manifest+json, per the media type registration MDN points at. Three sites send application/json and one sends application/octet-stream, and all four parse anyway.