Platform · We do not build on it
A hosted website builder where pages are assembled by dragging pre-made blocks, and the platform serves the result from its own CDN.
Tilda puts a page on the internet in an afternoon and then keeps it. The constraint is not the editor, which is genuinely good — it is that a Tilda-hosted site has no root an operator can write to and no API that writes back, so what the site publishes, and what a machine is permitted to fetch from it, are both decided by the vendor rather than by the person who owns the business. The refusal below is narrow and technical, and it inverts cleanly: Tilda is the wrong substrate for anything that must be read by machines at scale and still owned in five years, and the right one for something that will be taken down before Christmas.
01
Crawl policy is not a decision on Tilda, it is a default. Tilda's own answer base states that the robots.txt file cannot be modified, and that the only route to changing it is exporting the project to other hosting. The generated file is the platform's; the single lever handed to the operator is a per-page checkbox labelled "Forbid search engines from indexing this page", which Tilda's own answer base describes as the disallow rule for robots.txt and offers no way to shape beyond adding or removing it.. That forecloses the entire class of work where what a machine may fetch is written deliberately — no Allow rule for a named agent, no distinction drawn between Googlebot-Extended, GPTBot, ClaudeBot and a scraper with no stated purpose, no crawl-delay aimed at the one client hammering a catalogue, no second Sitemap line pointing at a feed built by hand. The only available position is refusal of a whole page, which is the bluntest instrument in the file and the one that costs the most.
02
The prohibition is far broader than robots.txt, and this is the part that gets missed. Tilda's answer base states plainly that files cannot be uploaded to a Tilda site at all, and directs operators to Google Drive or Dropbox instead. A Tilda-hosted site therefore has no writable root of any kind. There is no /llms.txt, no /.well-known/security.txt, no ads.txt, no CSV or JSON twin of a dataset served from the same origin as the page describing it, no Markdown rendition of a page under a stable path, and no arbitrary verification file — and no arbitrary verification file — Tilda's own instructions for proving Search Console ownership offer a meta tag or a DNS TXT record and never the HTML-file method, which is the single method that would require writing a file to the root. Every one of those is a file at a URL, and the platform's position is that a Tilda site has only the files Tilda generated.
03
The published API is seven methods and all seven begin with get: getprojectslist, getprojectinfo, getpageslist, getpage, getpagefull, getpageexport, getpagefullexport, capped at 150 requests per hour. Content can be pulled out. Nothing can be pushed in. Pages exist because a human dragged blocks into place, and they change because a human dragged them again. That is entirely tolerable for twelve pages and fatal at the point where a site must answer a hundred specific questions — every licence class, every model in a fleet, every documented failure mode of a product — because those pages are properly generated from a source of record and regenerated when the record moves. On Tilda that regeneration is a person and a mouse, repeated a hundred times, and repeated again when one price changes.
04
There is a hard ceiling of 1,000 pages per project which Tilda states cannot be raised, and the documented workaround is to buy the Business plan and split the excess across additional projects on separate subdomains. That workaround is worse than the ceiling. Splitting a corpus across origins fragments exactly what a knowledge graph is built to hold together — one entity, one canonical set of identifiers, one internal link structure that an extractor can walk without deciding whether two subdomains are the same organisation. The ceiling is reachable; the escape from it damages the thing the corpus exists to do.
05
The weight is structural rather than the result of a careless build, which is why it cannot be optimised away. Tilda's own English homepage measured 470,012 bytes of HTML uncompressed and 112,617 over the wire, plus thirty-four separate CSS and JavaScript files totalling 784,566 bytes uncompressed and 247,635 over the wire, before a single content image.. An editorial article page on Tilda's own tilda.education — https://tilda.education/articles-icons — measured 565,268 bytes of HTML, 1,676 div elements, 624 inline style attributes and 96 script tags. The block framework, its slider, its video processor and a 78 KB polyfill bundle ship on pages that plainly use none of them — a stock Tilda starter page at start.tilda.ws pulls all four — and there is no build step at which to drop them., and there is no build step at which to drop them. On the audience question the honest figure is smaller than the one usually quoted: StatCounter put the UAE at 52.62 per cent mobile against 46.8 per cent desktop in August 2026, On the audience question the figure depends entirely on what is being counted: StatCounter, which counts page views, put the UAE at 52.62 per cent mobile against 46.8 per cent desktop in August 2026, while the seventy-five-per-cent-and-up figures that circulate in regional decks come from survey data on which device people say they use, which is a different measurement and a larger number.. Half an audience is still half, and that half pays for the framework before it reads a sentence.
06
Tilda serves jQuery 1.10.2 — a July 2013 release, 93,064 bytes uncompressed — from its CDN to pages that use it. Every jQuery release below 3.5.0 falls inside the ranges of CVE-2020-11022 and CVE-2020-11023, and below 3.4.0 inside CVE-2019-11358. Whether any of those is exploitable on a specific Tilda page depends on what that page does with untrusted HTML, and this entry does not claim an exploit exists; what it claims is narrower and sufficient. A dependency more than a decade old sits inside the served page, the operator cannot upgrade the copy Tilda serves, cannot fork it and cannot patch it — a later jQuery loaded from the head zone sits alongside it rather than replacing it, and breaks the block library that binds to the old global, and cannot answer a security questionnaire about it with anything other than a request to ask the vendor. For a site that publishes a brochure that is an accepted risk. For one that carries a login, a cart or a form collecting anything a regulator cares about, it is a position with no remediation path.
07
Structured data on Tilda is automatic for exactly one thing: microdata on products in the catalogue. Everything else is a JSON-LD block pasted by hand into the head zone of one page at a time, which is precisely what Tilda's own instructions describe, down to recommending an external generator to produce it. Tilda's own English homepage, measured for this entry, carried zero JSON-LD blocks and not one <main>, <article>, <section> or <nav> element against 789 div elements — the document is a deep stack of numbered divs, which is the hardest possible input for anything trying to work out where the content begins. — the document is a deep stack of numbered divs, which is the hardest possible input for anything trying to work out where the content begins. Hand-pasting scales to a dozen pages. It does not produce what an entity needs in order to be recognised and cited: a connected graph in which Organization, Person, Service and Article nodes reference one another by stable @id across every page, with a dangling reference caught before publication rather than discovered months later in a validator. Catching it requires a build gate, and Tilda has no build.
08
Export is the lock-in question and it answers badly. It is gated to the Business plan, it produces a static HTML snapshot with images, CSS and JavaScript, and it drops a specific list of things on the way out: Tilda CRM and Tilda Members cannot be exported at all, the product catalogue and feeds cannot be exported, direct links to individual products and posts stop working and become reachable only by clicking through a list, page and project password protection is disabled, and the free SSL certificate is disabled. Forms keep working only while a paid Tilda subscription continues, which means the exported site is still tethered to the vendor for the one function that produces revenue. Worse than any single item is the direction of travel: Tilda's guidance is to make changes in Tilda and export again, so the export is a one-way snapshot rather than a fork. Any edit made to the exported files is destroyed by the next export, which means the moment an engineer touches the exported code, Tilda is no longer the source of record and the visual editor the whole platform exists to provide has been abandoned. There is no gradual migration. There is a cliff.
09
Where Tilda is right, and it is right often: a launch page, a campaign microsite, a conference or event site, a single-page announcement, a landing page behind paid media that will be rewritten twice and retired in six weeks. For those, publication speed beats every property above, and Tilda's advantages are real ones. A marketer changes a headline and it is live in seconds with no deploy, no review, no engineer in the loop and no way to break the build, because there is no build. The block library covers the layouts a campaign page actually needs, Zero Block is a competent freeform editor rather than a toy, and the CDN and SSL are handled without anyone thinking about them. A short-lived page does not need an llms.txt, will never be asked to hold a thousand URLs, and has no five-year ownership question, because in five years it will not exist. Refusing Tilda for a campaign page on the grounds set out above would be an engineering argument applied to a problem that does not have one.
An entry that cannot point at something you can open, run or read does not compile. That is a property of the type, not a promise in a paragraph.
Invariant
This argument is about machine readability, programmatic growth and ownership over years, and it collapses the moment none of those are at stake — a launch page, an event site or a campaign landing page loses nothing by being on Tilda and gains a publication cycle measured in seconds. It is also not a migration recommendation. A Tilda site that already ranks, already converts and is not being asked to do anything a builder cannot do should be left alone: a rebuild is paid in full on day one and returns over quarters, and the honest answer for most small brochure sites is that the payback never arrives. The refusal applies specifically to Tilda-hosted sites; a project exported onto infrastructure the owner controls recovers robots.txt, arbitrary files and response headers, at the price of the visual editor and of every subsequent re-export, and for some teams that trade is wrong. Finally, the measurements here were taken on 4 September 2026 against Tilda's own pages. Tilda ships changes, and a reader who runs the commands and gets different numbers should trust the numbers rather than this page.