We spent three weeks on meta tags for a mini-program no search engine will ever crawl. Sitemap, canonical links, a description we wordsmithed over two lunches — the whole ritual. Then we opened the host app, typed the phrase people actually use for our product, and watched a two-week-old copy sit above us.

Here's the thing: there's no web index behind that box. The host isn't ranking pages, because there aren't pages to rank. It's matching a query against a name, a category, and a few signals about whether people came back. Same word, search, completely different machine. We'll keep this to that one box — not discovery in general — because the tactics for it are oddly specific.

The index isn't a web index

A crawler follows links, reads text, builds a graph. Inside a super app, what gets indexed is one structured record: name, category, short description, a configured keyword list, an icon, some screenshots. On top of that sit behavioural signals — open frequency, whether opens repeat, ratings, how fast people leave. That's the whole ranking surface. Nobody crawls your product detail screen, because from the outside it isn't a URL.

Two consequences teams keep missing. Nothing on your public website leaks in: backlinks, a good blog, a content hub earn you exactly zero position inside the host. And the reverse holds too — no screen inside the mini-program will ever appear on Google. If you've been quietly hoping doing some SEO would bring organic traffic to an in-app catalogue, it won't. The two systems share no cable.

So the question isn't how do we rank. It's what does this host even have to match against. Usually a dozen words of text and a category dropdown. That makes each of those words unusually expensive.

What you can actually tune

Start with the name: most weight, cheapest to change before launch. Use the word a customer says out loud, not the one your branding deck invented. We shipped a made-up two-syllable name once and couldn't work out why nobody found it — nobody was typing that, they were typing car wash near me. Names go through uniqueness and review anyway, so you can't stuff four keywords in there even if you wanted to.

The category decides which pool you're allowed into. Go broad to reach more people and you become a mediocre match in a huge pool instead of a strong match in the right one. Descriptions and keyword fields are a short configured list, not prose: three to five real queries, in the order people ask them. Screenshots don't move rank directly, but they move the tap, and the tap feeds behaviour signals — so a splash screen there is wasted inventory.

FieldWhat the host actually readsDo thisDon't
NameExact and fuzzy match against the query, plus a uniqueness checkUse the word people say out loud, keep it shortInvent a brand term nobody types, or stack four keywords
CategoryWhich result pool you're allowed intoPick the pool your real use case sits inChoose a broad category to reach more people
Description and keywordsA short list of configured terms, not a paragraphThree to five real queries, in the order people ask themPaste the homepage copy or repeat the name six times
ScreenshotsTap-through, which feeds back into behaviour signalsShow the one screen people came forLead with a splash screen or a feature grid
Opens and repeat useWhether people come back, rate it, or bailEarn it with the productBuy fake opens; hosts discount or penalise them

Notice the last row isn't something a marketer can edit. That's the honest boundary here: align the record with the query, then let the product do the rest.

The part nobody publishes

Now the tradeoffs, and they're not small. The algorithm isn't yours. No host publishes its weights, and the ones we've inferred have shifted more than once. A tactic that lifted a listing last quarter can quietly stop working, with no changelog to read. Don't bet the growth plan on guessing weights, and don't rename the product every time someone forwards a ranking rumour.

Keyword stuffing is the obvious trap; hosts actively defend against it. Cramming one term into the name, description and keyword fields reads as manipulation and can get the listing demoted or rejected at review. The duller risk is renaming itself. Once a name is out, old share cards, screenshots and word of mouth all point at the old string — change it and some loyal users suddenly can't find you. Treat the name as a brand decision, not an A/B variable you tweak on a Tuesday.

And one limit we just accept: Google will never index the screens inside the mini-program. Your catalogue and help pages carry no weight past the app's walls. If you need searchable content, build it on the open web as a separate thing; it isn't a side effect of the listing.

Start somewhere unglamorous: read a month of support tickets and write down the exact words people use when they can't find you. Put one in the name, drop the listing into the category they'd actually look under, then leave it alone for a few weeks. If the query and the record match and people still don't open it twice, that's not a search problem anymore — it's the product, and no keyword fixes that.

Sources