Author: gptadmin

  • How to use AI writing help with your own key, without losing your voice

    AI writing tools can save time on social posts, especially when the same message needs several versions for different networks. They can also flood your feed with generic copy. The difference is mostly in how you use them. This guide covers GPTpost’s AI assist, which runs on your own key, and a routine that keeps the human in charge.

    Step 1: add your own key

    GPTpost’s AI assist works with your OpenAI, Anthropic or Gemini key. Requests go to the model account you already chose and are billed there. We do not resell tokens, and we do not add a new AI data processor to your setup. For many teams that also makes the conversation with legal or IT shorter.

    Step 2: ask for something specific

    Vague prompts produce vague copy. Compare:

    • “Make this better” gives a slightly different version of the same text, often with more adjectives.
    • “Shorten this to fit Bluesky’s 300 characters, keep the example” gives a useful starting point.
    • “Give me three alternative first lines, each under ten words” helps when the opening feels flat.
    • “Rewrite the LinkedIn version for Mastodon, plainer, no marketing tone” respects the network’s norms.

    Step 3: edit like an editor

    Treat every suggestion as a draft from a fast but inexperienced writer. Check facts, numbers, names and links. Remove filler words. Restore phrases that sound like you. If the suggestion adds a claim you cannot support, delete it.

    Step 4: keep approvals on

    An AI-assisted post goes through the same approval workflow as any other. The reviewer sees the final text per network, not the prompt.

    Where AI helps and where it does not

    AI is good at trimming, rephrasing and producing variants. It is weaker at judgment: whether this is the right week for the post, whether the audience already heard it, or whether the tone fits the moment. Those decisions stay with people.

    Why we chose bring-your-own-key: AI writing help without a new data processor.

    https://oraglegpt.org/blog/bring-your-own-ai-key-social-media/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Why we treat Mastodon, Bluesky and Lemmy as first-class networks

    When people look at the networks GPTpost supports, they sometimes ask why we put so much care into Mastodon, Bluesky and Lemmy when they are smaller than the big commercial platforms. Here is our reasoning.

    Where the conversations are

    Over the past few years, many developer, open-source, research and public-interest communities have built a real presence on text-first networks. For a devtools company, an open project or a technical publication, the conversations there are often more valuable than reach elsewhere: people ask detailed questions, share useful links and remember who helped them.

    Different networks, different norms

    These networks are not interchangeable with each other, or with the big platforms.

    • Mastodon uses a 500-character default limit, hashtags are a main way to be discovered, and the culture values content warnings, alt text and plain conversation over engagement bait.
    • Bluesky has a 300-character limit, supports images with alt text, and has its own norms around custom feeds.
    • Lemmy is organised around communities, posts need titles, and link dumps without context are not welcome.

    Pasting a LinkedIn post into each of them ignores all of that.

    How GPTpost handles them

    • Each network gets its own version of the post in the same composer.
    • Lengths are checked when you schedule, not when a post fails at its publication time.
    • Lemmy posts carry their own title and body.
    • Bluesky posts can include up to four images, each with alt text.
    • All of them share the same calendar, approvals and API as every other network.

    Small does not mean minor

    We do not expect these networks to replace the larger ones for every brand. But for the communities that live there, being taken seriously matters, and a scheduler should make that easy rather than awkward.

    More on scheduling for Mastodon: schedule Mastodon posts from a team calendar.

    https://oraglegpt.org/use-cases/mastodon-scheduler/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Five social media habits worth dropping (and what to do instead)

    Most advice about social media is about adding things: a new network, a new format, a new tool. Removing a habit can be just as valuable, and it frees up time for what works. Here are five habits many teams have dropped without regret, and what to do instead. We would also like to hear yours.

    1. Posting every day no matter what

    Daily posting can work when you have something worth saying daily. When you do not, it produces filler, and filler trains your audience to scroll past. Instead: pick a sustainable rhythm, even two or three good posts a week, and keep it consistent.

    2. The hashtag wall

    A long block of generic tags at the end of every post rarely helps and makes posts harder to read. Instead: a few specific tags where the network uses them, written in CamelCase, and none where they add nothing.

    3. Reacting to every trend

    Trend-jacking is risky and usually forgettable. Instead: react only when you have a genuine angle, and spend the saved time on evergreen posts that stay useful for months.

    4. Identical text on every network

    A post written for LinkedIn rarely fits Bluesky’s 300 characters, and the norms on Mastodon differ again. Instead: one core message, a version per network. A scheduler that keeps the versions side by side, as GPTpost does, makes this quick.

    5. Checking metrics many times a day

    Constant checking feels productive and changes nothing. Instead: a short weekly review: which posts got replies or clicks, which got neither, one change for next week.

    Your turn

    Which habit did you drop this year, and did you notice any downside? Tell us on any of our profiles. More: oraglegpt.org.

  • “Just a quick post”: why short posts take longest, and how to speed them up

    There is a running joke among people who write for social media: nothing takes longer than a quick post. You start at nine, discover the text is too long for Bluesky, cut a sentence, lose the point, rewrite the point, and end up deleting a single comma to get from 301 to 300 characters. Twenty minutes, one short post.

    It is funny because it is true, and because there is a reason behind it.

    Why short is hard

    Long text lets you circle an idea. Short text forces you to know exactly what the idea is. Most of the time spent on a short post is not typing, it is deciding. A 300-character limit is a very honest editor: it shows you immediately whether you know what you want to say.

    Write the shortest version first

    The most useful habit we know is to reverse the usual order. Instead of writing the long LinkedIn version and squeezing it down, write the shortest version first: the one-sentence point plus the one detail that makes it interesting. Then expand for the networks that allow more room. Adding context is much faster than removing it.

    Three cuts that rarely hurt

    • The warm-up sentence. “We are excited to share that” can almost always go.
    • Double qualifiers. “Really quite useful” is “useful”.
    • The list of three when two will do.

    Let the tool count for you

    Checking limits by hand is how posts end up rejected at the scheduled time. In GPTpost each network has its own version of the post, and lengths are checked when you schedule, so you find out at nine o’clock, not at publication time.

    Accept the comma

    Sometimes the final cut really is one comma. That is fine. It still counts as a quick post.

    More about GPTpost: oraglegpt.org.

  • Which social network surprised you this year? How to spot the quiet winners

    Every year, most teams find that at least one network behaved differently than expected. Sometimes a big network delivers less than it used to. Sometimes a small one, added almost as an afterthought, produces the best conversations. We are asking our readers which network surprised them this year, and in the meantime, here is how to notice the quiet winners in your own data.

    Look past reach

    Reach and impressions are easy to compare and often misleading. A network with modest reach can still be your most valuable if people reply, ask questions or click through. Look at:

    • Replies with substance, not just emoji.
    • Clicks to the pages you care about, using consistent UTM tags or short links.
    • Repeat interactions from the same people over months.

    Compare effort, not just results

    A network that brings a few good conversations for twenty minutes a week may be a better investment than one that needs daily attention for a little more reach. Note roughly how much time each network takes.

    Check the format

    Sometimes the network did not change, the format did. A question posted on a Tuesday evening, a short how-to, or a behind-the-scenes post may work far better than announcements on the same network. Separate “this network works” from “this format works”.

    Our own observation

    For a product like ours, the text-first networks (Mastodon, Bluesky and Lemmy) consistently bring thoughtful replies from developers and open communities. That is not a rule for everyone, which is why we are asking.

    Use it for next year’s plan

    Once you have spotted a surprising winner, give it a deliberate share of next quarter’s plan: a recurring format, a posting slot, a clear goal. Then check again in three months.

    Which network surprised you? Tell us at hello@oraglegpt.org or reply on any of our profiles. More: oraglegpt.org.

  • Per-user vs per-workspace pricing for social media tools: when each makes sense

    Social media tools use different pricing models: per user, per connected channel, per workspace, or a mix. None is good or bad on its own. Which one works for you depends on how your team is built and how it grows. Here is a comparison using published numbers.

    An example of per-user pricing

    When we checked on 15 October 2026, Hootsuite listed its plans per user per month, billed annually: Standard at $99, Professional at $199 and Advanced at $399, with Enterprise priced on request. Plans change, so always check the vendor’s current page.

    When per-user pricing is fine

    Per-user pricing makes sense for a stable team where each person uses a lot of the product every day. Hootsuite is a full suite for larger social teams: publishing, inbox, listening and analytics. If your team uses all of that, you are paying for people who get value from many features.

    When it becomes painful

    Per-user pricing grows with every person who touches the tool, regardless of how much they use it. Agencies and small teams often have:

    • a junior who drafts a few posts a week,
    • a client who only approves,
    • a manager who looks at the calendar once a month,
    • a freelancer who helps during busy periods.

    Each one can cost a full seat. The result is that teams share logins (a security problem) or leave people out of the tool (a workflow problem).

    The per-workspace alternative

    GPTpost is priced per workspace: Solo USD 12 (1 seat), Team USD 39 (5 seats) and Agency USD 99 (25 seats) per month. Client reviewers using review links need no seat at all. The trade-off is scope: GPTpost focuses on planning, approvals and publishing, without a social inbox or listening.

    How to decide

    1. List everyone who touches posts and what they actually do.
    2. List the features you use weekly, not the ones you might use.
    3. Calculate the yearly cost under each model for your real team.

    Our side-by-side page: GPTpost vs Hootsuite.

    https://oraglegpt.org/compare/hootsuite/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • How to see your social media schedule in Google Calendar, Outlook or Apple Calendar

    Most teams plan their social posts in one tool and everything else in their calendar. The two rarely meet, and that is how a product announcement ends up scheduled on the same morning as a company event, or a cheerful post lands on a public holiday nobody thought about.

    ICS export in GPTpost

    GPTpost can export your planned schedule as an .ics file, the standard calendar format. Every major calendar app opens it: Google Calendar, Outlook and Apple Calendar. Each scheduled post appears as an event at its planned time, with the post title and the networks it goes to.

    How to use it

    1. In GPTpost, export the schedule. You can choose a date window, for example the next four weeks, or export everything planned.
    2. Import the file into your calendar. Many people use a separate calendar for it, so it can be shown or hidden with one click.
    3. Look for clashes: launches, events, holidays, other campaigns.
    4. Adjust the plan in GPTpost, then export again if you want an updated view.

    Who finds it useful

    • Managers who want to see the plan without learning another tool.
    • Clients who live in Outlook and prefer to see the month at a glance.
    • Teams preparing a weekly content meeting.

    What it is not

    The export is a one-time download, not a live subscription URL. If the plan changes, the file in your calendar does not update by itself. We chose not to offer a public subscription link in this version because it needs a properly revocable design, and we would rather ship that carefully than quickly. For now, export again after significant changes.

    A simple routine

    Export next month’s schedule during your monthly planning session, import it into a shared “Social” calendar, and review it alongside the company calendar. Ten minutes, and fewer surprises.

    More about GPTpost: oraglegpt.org.

  • post_final_v3_REAL: versions per network vs versions per panic

    Open almost any shared marketing folder and you will find a sequence like this: post_final.docx, post_final_v2.docx, post_final_v3_REAL.docx, and finally post_final_v3_REAL_use_this_one.docx. It is funny until the wrong one gets published.

    Two kinds of versions

    Not all versions are a problem. It helps to separate two kinds.

    • Versions per network are good practice. Bluesky allows 300 characters, Mastodon 500 by default, LinkedIn much more. A blog or DEV article needs a title. Writing a version for each network is how one idea fits each place properly.
    • Versions per round of feedback are the dangerous kind. Each round creates a new file or a new email attachment. Nobody is sure which one is current, and comments from one round get lost in the next.

    How the wrong text gets published

    The pattern is familiar: the client edits v3, someone else edits v3 at the same time, the final approval refers to “the latest version”, and the scheduler holds a copy pasted in from v2. Every step was reasonable. The result is still wrong.

    A calmer setup

    1. One draft in one place. The post lives in the scheduler, not in a document next to it.
    2. Versions only where networks differ. The Bluesky cut and the LinkedIn text are versions of the same post, side by side.
    3. Feedback as comments on the draft. Nobody creates a new file to suggest a change.
    4. Approval on the exact text. The approver sees what will be published, per network, not a summary.

    GPTpost is built around that setup: per-network versions in one composer, comments on drafts, and approvals or client review links on the exact text.

    And the folder?

    Archive it. Keep post_final_v3_REAL_use_this_one.docx as a reminder, if you like. More about GPTpost: oraglegpt.org.

  • Scoped API keys for social media automation: how GPTpost limits access

    Connecting a script or an AI agent to your social accounts is useful and slightly nerve-racking. The useful part is obvious. The nerve-racking part comes down to two questions: what exactly can this key do, and what happens when circumstances change? Here is how GPTpost answers both.

    Scopes: only what the job needs

    Every workspace API key carries one or more scopes:

    • accounts:read: list connected social accounts, with network, handle, state and text limit.
    • posts:read: list posts and read their schedule and per-network publish status.
    • posts:write: create posts, schedule them and publish now.
    • media:write: upload media from a URL or as base64.

    A script that builds a weekly report needs only the read scopes. An agent that drafts posts needs posts:write, and perhaps media:write if it attaches images.

    Never more than its creator

    A key never has more rights than the person who created it. That prevents keys from becoming a quiet way to gain access. It also means keys follow people’s roles: if the creator loses a role or is removed from the workspace, their keys stop working. Changing or resetting their password, or replacing their second factor, revokes every key they created.

    Shown once, stored hashed

    A key is displayed only at creation. GPTpost keeps a hash, not the key itself. If a key is lost, revoke it and create a new one.

    Same rules as people

    Posts created with a key go through the same per-network text limits, approval workflows and plan limits as posts written in the app.

    Good habits

    1. One key per script or agent, named clearly.
    2. Minimum scopes.
    3. An expiry date.
    4. Revoke keys you no longer use.

    Full details: GPTpost social media API.

    https://oraglegpt.org/use-cases/social-media-api/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Why you should write hashtags in CamelCase

    CamelCase means capitalising the first letter of each word in a compound word, as in OpenSource or SocialMediaTips. For hashtags, it is one of the simplest accessibility improvements you can make, and it takes no extra time once it becomes a habit.

    The problem with lowercase hashtags

    Screen readers read text aloud for people who are blind or have low vision. When they meet a long lowercase hashtag, they often cannot tell where one word ends and the next begins. The result can be a garbled single word or letters read one by one. The reader loses the meaning, and sometimes the whole point of the post.

    Sighted readers are affected too. A long lowercase tag is harder to scan, and occasionally it reads as a different, unintended phrase.

    The fix

    Capitalise each word: SocialMediaTips instead of socialmediatips. Hashtags are not case sensitive on the major networks, so the tag still links to the same topic. Nothing is lost.

    Why it matters even more on Mastodon

    On Mastodon and the wider fediverse, there is no central algorithm deciding what people see. Hashtags are one of the main ways to discover posts, and many people follow tags directly. That makes them more visible and more important to get right. Fediverse communities also tend to care about accessibility, and CamelCase is widely expected.

    Making it a habit

    • Update any saved hashtag lists once. In GPTpost, hashtag groups let you store approved sets per client or topic, so fixing them once fixes every future post.
    • Add it to your brand voice guide as a one-line rule.
    • When reviewing drafts, check hashtags along with links and spelling.

    Related habits

    While you are at it: write alt text for images, avoid using emoji as bullet points in long lists, and put the important information in the text rather than only in an image.

    More on posting to Mastodon and Bluesky: how to schedule posts without losing what makes each network different.