Short version up front, because I hate articles that make you scroll for the payoff.
A 2022 version of this page listed ten "hidden gems" in Twitter and told you where to click for each one. Four years later the product has a different name, the company behind it has a different name, and that list has stopped being advice. Checked against current evidence, here is where those ten landed: five still work and cost nothing (muted words, polls, threads, Lists, notification filters), one survived but got harder to reach (advanced search, now effectively account-only), two are behind a subscription (the analytics dashboard, and anything resembling timeline customisation), one is gone (Moments), and one — scheduling — I can only verify in the form that got paywalled twice. Communities, a feature that didn't exist in 2022, has already been shut down too, on 6 May 2026.
That inventory is the article. Not because you need a list of buttons — because an old list of buttons is the only honest way to measure what a platform takes back. Every item below was once free, documented, and worth learning. Roughly half of them still are. The other half were withdrawn by a company that owed nobody an explanation and mostly didn't give one, and that is a fact about every product you build a habit on, not just this one.
I've argued the general case elsewhere on this site, in the pieces on patient software and on boring tech stacks. This is the same argument with receipts.
How I checked this, and what I couldn't check
Everything dated below comes from X's own current developer documentation or from dated reporting. What I could not do is open the product and look: help.x.com returns HTTP 403 to anything that isn't a logged-in browser, and the main site serves a JavaScript shell with no content until you sign in. Where the only available source was a search-engine-optimised blog with no date and no citation, I left the claim out rather than dress it up; those gaps are listed at the end.
One side door turned out to be very useful. The developer docs are not behind the wall, they carry dates, and they're written for people who will file a bug if they're wrong. If the API still bills you for reading a Space, Spaces exist. If the docs warn that the web search operator is spelled min_faves: and the API one min_likes:, web search operators exist.
The ten, one at a time
Muted words — still there, still free. No reported change since 2022, and muting is verifiably a live platform primitive: X added mute.mute and mute.unmute events to its Activity API on 21 July 2026, and its price list bills "Mute: Read" at $0.001 per record. Keyword muting is the most valuable thing in the original list and the least glamorous, which is probably why nobody has monetised it.
"Customize your home timeline" — gone, and then sold back. This is the clearest case in the inventory. In March 2022 Twitter shipped a sparkle icon that let you pin a reverse-chronological timeline. On 11 January 2023 that control was removed and replaced with the For You / Following tabs, with For You as the landing view and the app snapping back to it every time you reopened; after a month of complaints, on 6 February 2023 the mobile apps stopped forcing the switch, but no setting to pick a permanent default was ever added. Then on 22 April 2026 X shipped Custom Timelines: Grok classifies posts across more than 75 topics, you pin up to ten topics or Lists next to For You and Following, and there's a 24-hour snooze for a topic you're sick of. It is a genuinely better version of the thing. It is iOS-first and it requires a paid tier.
Free control, withdrawn, four years of no control, paid control. If you want one sentence for why I don't invest much in learning a product's chrome, that's it.
What the paid tier costs has also been a moving target: Premium+ went from $22 to $40 a month on 18 February 2025, and on 12 October 2026 X announced a new lineup that bundles X Premium with Grok and Cursor across Free / Plus $8 / Premium $30 / Super $100 / Ultra $200, with existing subscribers keeping their plans and rollout proceeding by region. Three different tier structures in four years, for the same product.
Scheduling — verifiable only in the form that was paywalled twice. TweetDeck was how most people scheduled. X gave 30 days' notice in July 2023 and made it subscriber-only on 15–16 August 2023, renaming it X Pro. Then on 26 March 2026 X Pro was pulled out of the Premium tier and restricted to Premium+ at $40 a month — no advance notice, mid-billing-cycle, so people who had paid for the month lost the tool they'd paid for. The clock icon in the composer is a separate question that the public sources answer three different ways; see the note at the end.
Advanced search — survived, and it's still the best thing on the platform. More on this below; it earns its own section.
Notification filters — structurally intact, practically different. The quality filter and advanced filters (mute accounts you don't follow, default avatars, unconfirmed email or phone) are all still documented and none moved behind payment. What changed is underneath: Musk said on 20 March 2023 that replies would rank accounts you follow first, then verified accounts, then everyone else, and on 25 April 2023 that verified accounts were now prioritised. The filters still decide what you *don't* see. Who reaches you is now partly a function of who paid.
Polls — still there, still free. Up to four options, 25 characters each, five minutes to seven days. Some 2023 coverage listed polls among subscription perks; that never became a real restriction, and I'd ignore it.
Threads — still there, still free, and now the budget option. Self-threading in the composer is unchanged. What arrived alongside it is long-form: paid tiers can post up to 25,000 characters, and in June 2026 X added API endpoints for drafting and publishing Articles. Note the detail that actually matters if you post long: mentions past character 280 don't notify the person you mentioned. Threads, the free feature, remain the only way to reliably tag someone deep in a long piece of writing.
Lists — still there, still free, and more load-bearing than in 2022. Easiest item to verify, because X's own pricing page bills reads against /owned_lists, /followed_lists, /pinned_lists and /list_memberships at the reduced "Owned Reads" rate introduced on 20 April 2026, and Custom Timelines explicitly lets you pin Lists. A List is still the closest thing the platform has to a feed you control, which is exactly why it is the one recommendation from 2022 I'd repeat without qualification.
Moments — gone. Creation was removed from mobile in October 2018 and then ended for everyone; Twitter Support announced it on 7 December 2022, three weeks before the original version of this page went up, saying you could still see past Moments. Whether a /i/moments URL renders in 2026 I could not establish, and nobody appears to have written about it since, which tells you how much the feature mattered in the end.
Analytics — paywalled. The native dashboard was free for years. In the week before 13 June 2024 it started demanding a Premium signup, and the rebuilt version — impressions, engagement rate, profile visits, link clicks, adjustable date ranges, later an audience breakdown — shipped for subscribers only. Most of those metrics had been free the week before. Nothing was built that didn't exist; a door was added to a room you were already standing in.
Features that arrived after 2022 and have already gone
Worth recording, because it shortens the feedback loop. Communities launched after the original list, and on 23 April 2026 X announced it would shut on 6 May 2026, with the admin migration deadline later pushed to 30 May. Head of product Nikita Bier's reason was blunt: under 0.4% of users, but responsible for 80% of spam reports, financial scams and malware, and half the team's time in some weeks. He called it a "Temu version of subreddits." Admins were pointed at public joinable XChat group chats.
Here's the bit I'd flag. As of today X's own enterprise API index still lists a Communities API, and the live price list still bills "Community: Read" at $0.005 per record. Six months after the consumer feature was shut down, the documentation for the paid interface to it hasn't caught up — which is what you should expect from docs as a source of truth about what currently exists, and why I checked everything above twice.
Spaces, by contrast, verifiably still runs: the price list bills Space reads and the Activity API delivers a spaces.start event. Bookmarks are free; bookmark *folders* have been paid since Twitter Blue launched in June 2021, so if you read somewhere that folders used to be free, they didn't. DMs changed default in July 2023: unless you opt out, only people you follow and verified — meaning paying — accounts can message you. Encrypted DMs shipped in May 2023 requiring *both* parties to be subscribers, which is a strange thing to charge for.
Advanced search, which is the one worth your time
If you take one practical thing from this piece: learn the search operators and stop using the search box as a search box.
The operators still work. I can show that without a login, because X's API documentation — read today — carries a warning telling developers that the engagement operators are named min_likes: and min_reposts: in the API, and that "the equivalent operators on X web search are named min_faves: and min_retweets:", which will be rejected with a 400 if you send them to the API. You don't write that warning about a dead feature. The form at /search-advanced also still resolves — it returns 200, not a 404 — though like everything else on the site it needs a session and JavaScript before it shows you anything.
What's worth knowing, all of it still documented on the API side and all of it transferable to the web box: from: and to: for a single account, since: and until: to bracket a date range, url: to find everyone who linked a page, list: to search inside a List you control, conversation_id: to pull a whole thread, lang: to cut 50 languages down to one, - to exclude, has: for media, links and replies, and the engagement floors to skip the noise. Three or four of those combined will beat any amount of scrolling, and none of it costs anything.
Two honest caveats. First, search is effectively account-only now. Logged-out browsing was blocked outright on 30 June 2023, described at the time as a temporary emergency measure against scraping, and partially relaxed on 5 July 2023 — a link to a single post opens without an account, but profiles, replies and search do not. A capability available to anyone with a URL now requires a relationship with the company.
Second, search gets quietly re-tuned under you. On 19 January 2026 X deprecated the count-based operators on its search endpoints, saying it was "evaluating how to best re-enable these in the future." On 4 May 2026 it migrated search to a new index, restored those three operators as "new precision operators" — and in the same sentence noted that retweets are no longer returned in keyword search results at all. If you had built anything that counted reposts of a keyword, it silently started returning a different answer that day. Nothing broke. The numbers just changed meaning.
What actually happened underneath
The feature list is the symptom. The mechanism is the interface.
On the night of 12 January 2023 the major third-party clients — Tweetbot, Twitterrific, Echofon, Fenix, Talon — stopped working at once. No announcement, no warning, nothing on the developer forum. The first official word came five days later and was one sentence: "Twitter is enforcing its long-standing API rules. That may result in some apps not working." No rule was named. The rule arrived on 19 January, when the developer agreement was quietly amended to prohibit using the API to "create or attempt to create a substitute or similar service or product to the Twitter Applications." The terms were changed to retroactively describe what had already been done. Tweetbot, April 2011 to January 2023, and Twitterrific were pulled from the App Store the same day.
Then the API itself. On 1 February 2023 Twitter announced free access would end on 9 February; the date slipped to the 13th, then slipped again with no replacement date. What landed at the end of March was a free tier capped at 1,500 posts a month, a Basic tier at $100 a month with 10,000 reads, and an unpublished enterprise tier that *Wired* reported at around $42,000 a month. A $5,000-a-month Pro tier followed in May. The Academic Research track, introduced in January 2021 specifically to give researchers full-archive search and a 10-million-post cap, was discontinued along with the rest. A survey of 168 research groups presented the following year found roughly 60% of projects stalled, cancelled or moved to another platform, and — this is the part I find hard to read past — over half of the researchers surveyed also cancelled work out of fear of being sued.
Today it's metered. X piloted credit-based pricing in October 2025 and launched X API Pay-Per-Use on 6 February 2026: no subscription, buy credits, pay per request. The current list rates, from X's own page: $0.005 per post read, $0.010 per user read, $0.015 to create a post — and $0.200 to create a post containing a URL, thirteen times the cost of posting without a link. Self-serve accounts are capped at three million post reads per billing cycle and at 512-character queries, where enterprise gets 4,096. Along the way the Following, Likes and Quote-Posts endpoints were removed from every self-serve tier on 20 April 2026, and since 23 February 2026 you may only post a programmatic reply if the original author "summoned" you by mentioning or quoting your account.
To be fair about this: most of those changes have a defensible reason. The reply restriction is aimed at LLM-generated spam, which is real and getting worse. The link surcharge is aimed at link farms. Metered pricing is genuinely friendlier to a hobbyist than a $100 floor was. The point is not that each decision was wrong. The point is the shape: every one of them changed, with little notice, what your code was permitted to do, and at no stage was there anyone to appeal to.
The workarounds get taken away too
This is the part people skip when they reassure themselves that they'll just scrape it.
Nitter was the web front end that let you read a post without an account. In late January 2024 X disabled the guest-account mechanism it depended on and it died as its last tokens expired. It came back in February 2025 on a different approach. Then on 24 August 2026 X Corp sent cease-and-desist letters demanding that every instance and the project's repositories come down, giving until 5pm Eastern the following day and citing "unlawful use and circumvention of X's Application Programming Interface (API) and associated data." The repository was archived on 25 August and the best-known public instance went dark. On 6 September 2026 the project announced that, following legal advice, it would continue — so this one isn't over.
Killed by a server-side change, revived, then met with a letter from a legal department. Scraping is not a hedge against dependency. It's the same dependency with worse ergonomics and a lawyer attached.
What the people who used it said
I went looking for the arguments people had with each other at the time rather than the retrospectives. The best of it is on Hacker News. Reddit and Indie Hackers both return HTTP 403 to this machine, so I can't speak for what was said there.
The 2 February 2023 thread on the end of free API access opens with the comment that has aged best. "This is why I will never build a company based around a free API from a for-profit company," wrote one developer. "I'd never be able to sleep well at night knowing that years of work and all the employees/families that depend on the business is dependent on someone who could wake up one day and decide on a whim to end access." The first reply sharpened it, and I think improved it: paid changes nothing. "The existence of a paid API does not automatically guarantee any future pricing or even availability. One could build a company around a paid API only to have the pricing jacked up 10x or the API shut down or neutered in some critical way and end up in, functionally, the same situation."
The pushback in that thread is worth quoting too, because it's the obvious objection and it isn't stupid: "You should never build a company based around phone calls, because one day your phone might get cut off. You should never build a company around electricity, because one day your electricity might get cut off." I don't think it holds, and the reason is the whole distinction I care about. Electricity is a metered commodity with regulators, published tariffs and contracts. The X API is one company's unilateral permission, revised by announcement. The 12 January cutoff, the terms rewritten a week later to cover it, the endpoints removed on 20 April 2026 — none of those involved anything you could appeal.
The people who lost something were specific about what. A long-time third-party client user: "I'd even have happily paid Twitter to continue using Tweetbot. But this has destroyed the trust and the user experience of many of the power users who are most engaged with Twitter." A hobbyist in the February thread: "I have some personal Twitter automation scripts that only make 20 requests per week or so. No way I could justify paying for API access for something small like that."
And the one I keep coming back to, because it's the cost landing where it does the most damage. A blind user wrote to the Twitterrific developers as they shut down: "Your iPhone app allowed me as a blind person to use Twitter so much better than the app that they themselves produce." The reply: "Maybe the best thing anyone's ever told us. Thank you for this, truly. It is everything." Accessibility work is the clearest case there is of a third party doing something the platform wouldn't, and it had no standing whatsoever when the decision came. The shutdown took eleven years of it on a Thursday night.
By March 2023 a developer was arguing the API was effectively unmaintained, on the evidence that twelve of the twenty newest posts on the official developer forum were people whose access requests had gone unanswered. Three years later it's documented and metered again and clearly has engineers on it — a real improvement, and also the third different regime in four years, which is the pattern rather than an exception to it.
What I'd actually do
- Learn the search operators, not the interface. They've survived every redesign and both ownership eras, they're documented, and the knowledge transfers to other search systems. Everything you learn about where a button lives is worth nothing the next time someone reorganises a menu.
- Use Lists. It's the only free mechanism on the platform that lets you decide what you see, it's verifiably still there, and it's now the unit that the paid timeline feature is built on.
- Set muted words once and properly. Highest-value, lowest-glamour feature in the original list, still free, and nobody has found a way to charge for it.
- Assume anything you can only do through one company's interface is on loan. Not out of cynicism — as a planning input. Four years took one feature to a paywall, removed two, killed a third before its first birthday, and rewrote the terms of the developer interface three times.
- If you're writing code against it, price the switch now — not whether you'd migrate, but what it would cost on the day you're told you have a week. In February 2023 people had seven days, and a lot of them found out from a forum thread. And don't treat scraping as the fallback: Nitter was the best-engineered answer to exactly this problem and it has been killed twice, the second time by a letter rather than a code change.
- Keep your own copy of anything you'd miss. Bookmarks, a List's membership, your own posts. The data you can export is the only part of this that nobody can reprice.
After thirty years of writing software, the thing I notice isn't that platforms change — of course they change. It's how much of what we call expertise turns out to be memorised trivia about someone else's menu, and how little of it survives a product decision made in an afternoon. The features from 2022 that are still useful are, almost without exception, the ones closest to being a language: operators, a saved set of accounts, a list of words you don't want to read. The ones that are gone were all features of a *place*. That's not a coincidence, and it's a reasonable way to decide what's worth learning next time.