Digital marketing

Instagram marketing for a small business

Instagram does one thing better than any other channel: putting you in front of someone who was not looking for you. In exchange, the audience you gather there is not yours, and whatever you build has to be kept somewhere else as well.

  • Lesson 3 of 8
  • Beginner
  • Free, no signup

The four Instagram surfaces and the job each one does

One piece takes a different shape in each column. The columns are not stages and their order is not binding.

  1. Post

    2
    • The thing that stays

      If it will still be useful in three months, this is where it belongs.

    • The answer to a real question

      A question someone actually asked you in the DM this month.

  2. Stories

    3
    • The daily rhythm

      It lives twenty four hours, and that is what frees it.

    • Questions and polls

      The one place the audience answers at no cost to itself.

    • Live, when there is something to say

      Not because a live is due this week.

  3. Reels

    2
    • Reaching past your followers

      The only surface that does this systematically.

    • The opening seconds are everything

      Someone who does not know you will not wait through an intro.

  4. The DM

    2
    • Where the buying happens

      And the surface that gets the least attention.

    • An answer the same day

      The simplest move nobody sells a course about.

Live did not get its own column because it is not a separate surface; it sits inside the stories rhythm. There is no number here about the reach of these surfaces, because we have not measured it and Meta has not published it.

Last checked: Facts and tool names in this lesson are re-checked against their sources on this date.

What Instagram does, and what it does not do

Someone is scrolling with no goal, an image stops them, and your name enters their head without them searching for anything. No other channel does that as cheaply. Google waits for a person to have a question; Instagram does not wait.

Now the other side. The audience you gather there is not yours. You have no phone number, no email, and if the account is locked tomorrow you are left with no list at all. That is what we called a rented audience in the first lesson of this path: there is access, there is no ownership.

There is also one technical limit that changes the whole strategy: a link inside a caption is not clickable. Every route out of Instagram passes through one point, the link in the bio, and whatever you put there has to make that single click worth it.

The practical consequence: take Instagram for being seen and for building first trust, not as an archive. What you explained there will not be findable next year; the same explanation on your own site will be.

A square glass frame standing on loose sand with a red tall strip, green grid tiles and a blue circling reel

What each Instagram surface is actually for

You have four surfaces and each does a different job. The common mistake is publishing one piece across all four in the same shape.

A post is the thing that stays and gets saved. If a piece will still be useful to someone in three months, that is where it belongs. Stories are the daily rhythm and live for twenty four hours; their real value is not the view, it is the question and the poll, because this is the one place where the audience answers you at no cost to them. A live session is part of that same rhythm rather than a separate channel: it means something when there is something to say, not because a live is due this week.

Reels are the only surface that systematically reaches people outside your followers. And the DM is where the buying actually happens here, which is exactly the surface that gets the least attention.

Our position is blunt: if you only have time for one thing, take the DM seriously rather than the reel. A post that is seen and whose DM goes unanswered has only cost you the making of it.

How Instagram decides what to show

The only citable source here is Meta itself, and Meta has published this. On the official Instagram Ranking Explained page, four families of signal are named for Feed: your own activity, information about the post, information about the person who posted it, and your history of interacting with them. For Explore it is stated more bluntly: how many people, and how quickly, are liking, commenting, sharing and saving a post.

Two things that are not on that page matter as much as the things that are. First, no weighting is published; Meta does not say a save is worth several likes. Second, hashtags are not named in that signal list at all. That does not prove hashtags do nothing, but it does mean anyone handing you a precise number for hashtag weight is quoting a source that does not exist.

On that same page Meta also rejects a common claim, writing that it does not suppress content to get people to buy ads, and for that question it built a tool called Account Status which says why your content may not be recommended and lets you appeal. If reach is genuinely worrying you, that screen is the right place to look rather than a rumour thread.

Our position: stop hunting for a formula nobody published. All four signal families come back to one thing in practice, whether a real person did something deliberate with your post. That is the only lever you actually hold.

The four signal families Meta itself names for Feed

These four come from Meta's own published page, not from a tutorial video.

  • Your own activity

    Posts you have liked, shared, saved or commented on. Which means one person's feed differs from the next.

  • Information about the post

    How many people, and how quickly, liked, commented, shared and saved it, plus when and where.

  • Information about who posted

    How many times people have interacted with that account in recent weeks. A new account starts from zero, and that is normal.

  • Your history with that account

    Whether you comment on each other's posts. A two way relationship, not a one way follow.

Their order is not a weighting: Meta publishes no weights and we are not inventing any. Hashtags are not named in that list at all, which means neither that they do nothing nor that they work.

Why a save and a send beat a like

First the boundary: Meta names those four actions together and nowhere says which one counts for more. The order in this section is what we observe in day to day work, not a quotation from Meta.

Now the reasoning. A like costs nothing and ends there. A save means somebody said I will need this later, which is intent; a person who saves comes back. A send to a friend outranks both, because it puts your post in front of someone who does not follow you and whom you did not pay for.

So the kind of post changes. Make posts worth keeping rather than worth liking: a checklist, a table of sizes and prices, the order in which the steps have to happen, a comparison of the two options a customer genuinely gets stuck between. Those get saved. A pretty picture with nothing in it gets a like, and that is the end of it.

We use one simple test inside our own team: if nobody would screenshot this post to send it to a friend, the post is not ready. That test is more reliable than any percentage you hear about the algorithm, because you can answer it yourself before publishing.

Buying followers, and why the answer is no

Meta's spam policy states it plainly: selling, buying or exchanging engagement, meaning likes, shares, views, follows and clicks, is prohibited, and so is selling or buying platform assets such as accounts and pages. So before any marketing argument, this breaks the rules of the very place your page lives in.

The practical reason matters more. Those accounts do not buy anything. Worse, they ruin the only number that is genuinely yours. You have no standard figure for the Iranian market, and your only valid comparison is against your own previous month; once a few thousand lifeless accounts join your followers, you can no longer tell a weak post from a fake audience. You broke your own instrument to display a bigger number.

Mass follow and unfollow, and engagement pods, are the same trade in different packaging. The signal Meta counts is a deliberate action by a real person, and you are diluting exactly that.

One boundary is worth drawing clearly, because it often gets blurred: paying a real page with a real audience to show your post is not buying followers; it is advertising, and there is nothing wrong with it, provided the audience is real and the fact that it is an ad is not hidden. If that is what you want, we run the market for it ourselves: on the Instagram advertising market the page owner posts the price for a post or a story directly, and our commission is zero.

Bought followers against an audience that came on its own

Bought followers

  • Meta's spam policy prohibits buying engagement and follows
  • it does not buy and it does not message you
  • it makes the comparison with your own previous month meaningless
  • it dilutes the very signal Meta counts

An audience that came on its own

  • it arrived by a decision and can be reached again
  • it saves and it sends, which is the deliberate action
  • your month over month number means something again
  • there is less of it, and that little is reliable

This figure is about buying followers, not about paying a real page to show your post; that is advertising and the body of this lesson separates the two.

A week you can actually keep

Most business pages do not die because the content is bad; they die because the schedule was one only the first two weeks could sustain. Three posts a week that continue for a year beat ten a week that stop after a month, and we did not get that from any algorithm; we got it from the fact that your audience is whoever still remembers you exist.

Before any post, write one page and finish it: who you are writing for, what exactly you sell, what you will never say, the five questions a customer actually asked you this month (copy them from your DMs, do not invent them), and the proof you own. That page is the asset; posts are consumable.

Then one simple weekly loop: plan, batch, publish, answer messages, read your own numbers. Answering a DM the same day is more effective than every free tactic being sold in Persian, and that is exactly why nobody sells a course about it.

And measurement: compare against your own previous month, not a number you heard in a video. If you want to know where along the route a customer is lost, the marketing funnel lesson counts precisely that, step by step.

The weekly loop, five steps and no more

The fifth step is the one everyone drops, and then nobody knows what worked.

One weekthat can be repeated fifty times
  • 1 Plan from that one page
  • 2 Make in a batch
  • 3 Publish
  • 4 Answer the DMs
  • 5 Read your own numbers

A number of posts is deliberately absent from this loop. The right week is the one you can hold for a year, even if it is smaller than somebody else's.

The fast path, with AI

The usual way is sitting down every day and wondering what to post today. The fast way is writing your positioning down once, then letting one batch pass produce thirty concepts for a month that you choose from. The thing that separates this from the usual advice: do not ask for captions in the first pass. Concepts only. A concept can be rejected in ten seconds, a caption cannot, and that is where the speed comes from.

  1. Write a one page brief and finish it there: for whom, what exactly you sell, what you will never say, the five questions asked in your DMs this month (copy them, do not invent them), and the proof you own, such as photos of past work or a before and after.
  2. A cheap fast model of the Flash class is enough for this pass, because its work is recombining what you supplied. The current pick is listed in the AI section of this site. Keep the expensive model for the second pass, where you have to judge which concept is right for you.
  3. Run the recipe below. Cross out on the spot any concept you cannot photograph or film with what you have this week, even a good one. A concept that never gets made is not a concept.
  4. Keep eight to twelve of the thirty and throw the rest away. A high throwaway rate is itself the sign that this is working; if you kept twenty five, the brief was not written strictly enough.
  5. Write or rewrite the captions yourself, and copy no price, specification or claim about your product out of the model output. Voice and numbers are the two things the model does not get to decide for you.

Copy-ready recipe

I run a {type of business} and I write for {precise audience}.

What I sell: {two lines}
What I never say: {for example, promising a guaranteed result}
Five real customer questions from this month:
1) {question}  2) {question}  3) {question}  4) {question}  5) {question}
The proof I own: {photos of past work, before and after, samples, ...}
What I have this week: {only a phone / daylight / ...}

Rules:
- Do not write captions. Concepts only, one line each.
- Write no number, price or claim that I did not give you.
- Propose no concept that cannot be made with the equipment above.

1) Give me thirty post concepts, each with three things: the concept in one line, which question above it came from, and its shape (post, reel or story).
2) Beside each one, write why somebody would save it or send it to a friend. If you have no reason, delete that concept yourself.
3) Order them so they cover a month and two similar concepts never land back to back.
4) At the end, say which customer question is still unanswered.

Before you trust the output: Check three things yourself before trusting the output. First the voice: the concepts a model produces often carry the tone of a foreign brand and land as artificial in Persian; the final caption has to be written in the sentences you actually use with a customer in the DM. Second numbers and claims: every price, every specification and every sentence about a result has to come from you and not from the model. And third, reality: a concept attached to none of the five customer questions is a guess, however good it looks.

AI in this kind of work

For Instagram work, AI genuinely helps in three places and is dangerous in one. It helps turning what you already know into a month of concepts, turning one piece into the shapes each surface wants, and reading your own numbers. It turns dangerous where it makes the image itself and nobody checks what has been published.

Tools that actually help

  • Claude It holds a constraint like no captions, concepts only across a long output, and imitates a voice better when you give it samples. Iran is on neither of Anthropic's two supported country lists; we read that on Anthropic's own page rather than measuring it.
  • Gemini It reads Persian natively, so a Persian brief needs no translation step first. It also generates images, which is where the watermark question begins. Google's own page says the app runs in over 230 countries and territories, and Iran is not on that list.
  • ChatGPT It is a common pick for this same batch job. We write no claim about access from Iran: OpenAI's supported countries page returns a 403 to our server, and not writing a claim with nothing behind it is part of the job.

Where it backfires

The main risk here is the image, not the text. Google's image models mark their output with SynthID, an imperceptible watermark that Google says it can detect and that cropping and compression do not simply remove. So a page whose images are all model output is identifiable to anyone holding the detector, however natural it looks to a human eye. This is not a prohibition; it means knowing what you publish, and if you have a real product, a real photograph of it almost always sells better. The second risk is quieter: customer DMs are personal data, and before pasting them into a chatbot, read that vendor's data usage page, such as Anthropic's page on training.

Sources: Google DeepMind: SynthID Anthropic: is my data used for model training Anthropic: supported countries and regions Google: where Gemini Apps are available

Where this advice stops

Three things are deliberately absent from this lesson. First, any number about Instagram itself: no best posting time, no posts per week, no signal weights. Meta publishes no weights, and any number we invented would be wrong by next month. Second, we make no claim about access to Instagram from inside Iran; that situation changes and we have no citable source to write it from. Third, Meta's own paid advertising is not covered here, because we found no published, citable sentence about whether an Iranian business is eligible for a Meta ad account; the shape of that same question, with an official source, is in the lesson on Google Ads from Iran further along this path.

From our own work

On Instagram almost everything is an image, and the place where automation quietly breaks in Persian is exactly that: writing Persian text inside a picture. We build the share cards for our proposals on this very server, and the route is written down in inc/proposal-cover-render.php, where we also explain why GD was dropped: GD does not join Persian letters. The route that works is Imagick with the pango layout engine, because it does the shaping and the right to left work itself. That is why the Peyda font was converted from woff2 to ttf and three of its weights are installed in /usr/local/share/fonts/rgb/, and why rgb_prop_can_render_cover() asks Imagick::queryFormats('PANGO') before anything else and refuses to start when the answer is empty. Checked again on this server today: the imagick extension is loaded, PANGO is in the format list, and all three font files are in place. If you ever automate the text images for your page, that is where your time goes, not into the design.

Real follow-up questions

How many posts a week should I publish?

We have no published number to give you, and the numbers you hear have no source either. The right number is the one you can hold for three months without a break; if in doubt, start at your minimum and add after a month. Small and regular beats large and interrupted.

Do hashtags still work?

On Meta's own ranking page, hashtags are not named among the signals at all. That neither proves they do nothing nor proves they work, but it does mean nobody can tell you that thirty hashtags beat five. Treat them as topic labels, not as a strategy.

Where do I put my site link so it gets clicked?

A link in a caption is not clickable, so in practice only the bio is left, and it has to point at the newest thing you are promoting. One detail that is often skipped: put a campaign tag on that link so your analytics can tell which post brought the visit, otherwise all the traffic arrives as one shapeless lump.

Keep learning