Table of Content
I asked three different AI tools to write an Instagram bio and a hashtag set for a small ceramics studio. All three came back in under ten seconds. All three were fluent. All three gave me twenty-eight hashtags.
Instagram had capped hashtags at five months earlier.
One bio ran to 168 characters, which Instagram will not accept. One suggested the handle @clay.and.co_. , which ends in a period, which Instagram also will not accept. And one bio said the studio had been "featured in leading design press," which was news to the studio.
None of the tools were broken. They were doing exactly what they do: producing text that looks like good advice, drawn from an internet where 30 hashtags was gospel and nobody had told them otherwise.
That is the whole problem in one paragraph. AI is excellent at the shape of a bio and terrible at the state of a platform. So here is the process I run on anything a generator produces before it touches a real profile. It is for creators, small brands, social managers, and anyone who has ever pasted a hashtag block without reading it.
Three things changed while your tools were not looking
Instagram cut the hashtag limit to five. The change was announced in December 2025 and rolled out gradually. In the app, adding a sixth tag now triggers a warning and is blocked. Adam Mosseri has said repeatedly that hashtags do not increase reach; they help the system categorise a post and support search. The awkward part: Meta's own scheduling tools were still accepting up to 30 well after the app stopped, and a lot of "2026 hashtag limits" articles still say 30. If a human writer can get that wrong, a model trained on those articles absolutely will.
Hashtags stopped being a growth lever. This matters more than the number. For a decade the advice was volume: stack broad tags, hope for Explore. Instagram now classifies a post from the caption, the visuals, the audio and what people actually do with it. Five tags is not a smaller version of the old strategy. It is a different job.
Disclosure became enforceable, and your bio does not count as one. The FTC's Endorsement Guides, revised in 2023, say a disclosure on social media must be unavoidable: not hidden behind a "more" link, not mixed into a block of hashtags, and not sitting only on your profile or About page. Separately, Article 50 of the EU AI Act starts applying on 2 August 2026 and requires disclosure of AI-generated text published to inform the public on matters of public interest, unless a named person has genuinely reviewed it and taken editorial responsibility.
MARGIN NOTE
The single most useful sentence in the FTC guidance, for anyone in this niche: a disclosure that lives only in your bio is not a disclosure. People do not read your bio before they read your caption. If a post is sponsored, the post says so, above the fold, in plain words.
The five ways AI profile copy fails
Bad AI output is rarely bad in an interesting way. It fails along the same five seams, whether it is writing a 150 character bio or a 30 word caption hook.
| FAILURE MODE | HOW IT SHOWS UP | WHO CATCHES IT | COST IF MISSED |
|---|---|---|---|
| Invented credibility (severe) | "Featured in Forbes." "Award-winning." Follower counts, press mentions and certifications nobody can produce. | The one follower who checks | Trust, and a screenshot that outlives the bio |
| Expired specs (severe) | 30 hashtags. Wrong character counts. Advice about following hashtags, a feature that no longer exists. | The app, when it refuses to post | Failed posts, suppressed reach, wasted strategy |
| Invalid handles (high) | Hyphens, spaces, trailing periods, 34 characters, or a handle already taken on three platforms. | The signup form | An afternoon lost, or a split brand across platforms |
| Copy-paste sameness (high) | Every bio: "Helping X achieve Y." Sparkle emoji. "DM for collabs." The identical five tags on every post. | Bored humans, and spam filters | No follows, and a spammy-behaviour flag |
| Buried disclosure (severe) | #ad at the end of a tag block, below the "more" cut, or only in the bio. | Regulators, eventually | Liability for you and the brand that hired you |
Severity reflects how hard the failure is to reverse once it is live, not how often it happens.
Verify the claims, then verify the specs
A language model does not look up a fact and describe it. It produces the text that a correct answer would plausibly look like. Usually those coincide. When they do not, nothing inside the model flinches. It writes the wrong character limit in exactly the same confident tone as the right one.
So confidence tells you nothing, and there are two separate things to check.
The claims about you. Anything in a bio that a stranger could disprove: press mentions, awards, client names, follower milestones, credentials, "trusted by 10,000 creators." If you cannot produce the receipt in ten seconds, it comes out. Models add this kind of flattery unprompted because bios in their training data are full of it.
The claims about the platform. Every limit below was checked on the date at the top of this page. Check them again before you rely on them, because that is the entire lesson here.
| FIELD | LIMIT | WHAT ACTUALLY BITES YOU |
|---|---|---|
| Instagram bio | 150 characters | Spaces, line breaks and emoji all count. Emoji often count as two. |
| Instagram username | 30 characters | Letters, numbers, periods and underscores only. No leading, trailing or doubled periods. |
| Instagram display name | 30 characters | Separate from the handle, and weighted in search. Put a keyword here. |
| Instagram caption | 2,200 characters | Only about the first 125 show before the "more" cut. That is your real limit. |
| Instagram hashtags | 5 per post or reel | Down from 30. Schedulers may still accept more; the ranking system will not use them. |
| TikTok bio | 160 characters | Recently doubled. Most guides still say 80. |
Verified 10 July 2026 against platform behaviour and current reporting. Treat any spec table, including this one, as perishable.
Every claim in the bio has a receipt, and every number came from the app rather than the model.
Pressure-test the handle before you fall in love with it
A generated username has to survive four tests, and most fail at least one. Run them in this order, because there is no point checking availability on a handle the form will reject anyway.
▪ Legal characters. Instagram allows letters, numbers, periods and underscores. No spaces, no hyphens, no ampersands. A period cannot start or end the handle, and two cannot sit side by side. AI loves a decorative trailing dot.
▪ Length. Thirty characters is the ceiling. Ten or fewer is what people can actually type from memory.
▪ The telephone test. Say the handle out loud to someone and ask them to type it. This is where rn reads as m, capital I reads as l, and underscores vanish. If they get it wrong, so will everyone tagging you.
▪ Cross-platform availability. Check every platform you might want in two years, not just the one you are opening today. A handle that is free on Instagram and taken on TikTok is a branding decision, not a coincidence.
One more, which is not a rule but a scar: if you are changing an existing handle, announce it first. Post it, put "formerly @oldhandle" in the bio for a few weeks, and update your other profiles the same day. Old links and old tags do not follow you.
A stranger typed the handle correctly from hearing it once, and you own it on every platform you care about.
Treat the hashtags as labels, not fuel
This is where AI output is most confidently wrong, because the internet it learned from is full of hashtag advice from 2018. Delete the whole block and start again with five slots.
The mental shift: a hashtag no longer buys you reach. It tells the system what the post is about so it can be found later, in search, by people looking for exactly that thing. You are labelling a shelf, not buying an ad.
| SLOT | ITS JOB | EXAMPLE FOR A CERAMICS STUDIO |
|---|---|---|
| 1. Category | Tells the system the broad subject | #ceramics |
| 2. Niche | Narrows it to your actual corner | #wheelthrownpottery |
| 3. Place | Where searchers are looking for you | #brightonmakers |
| 4. Intent | What the viewer wants to do | #potteryclasses |
| 5. Branded | Collects your own work over time | #claystudiodiary |
Three rules the generator will not tell you. Hashtags count toward the caption character limit, so a wall of them eats your hook. Rotate your sets rather than pasting an identical five under every post, because identical repetition reads as automation. And check tags against banned and restricted lists before use, because a single suppressed tag can quietly limit a post's distribution.
You have five tags, each doing a different job, and you can say out loud why each one is there.
Read it aloud, then delete the machine
One hundred and fifty characters is not enough room to sound like everybody else. Read the bio out loud. If it sounds like a LinkedIn headline wearing a sparkle emoji, it came from a model and it will be forgotten in a scroll.
| THE TELL | WHAT IT SOUNDS LIKE | THE FIX |
|---|---|---|
| The helping formula | "Helping busy founders scale their brand." True of nine thousand accounts. | Say what you make, for whom, and one strange specific. "Wheel-thrown mugs. Mostly blue. Brighton." |
| The emoji bullet | A stack of arrows, sparkles and checkmarks doing the work words should do. | Keep one emoji if it earns its two characters. Delete the rest. |
| The compulsory triad | Everything arrives in threes: clear, concise, compelling. | Cut one. Two is usually the true number, and asymmetry sounds human. |
| The buried hook | The caption spends its first 125 characters clearing its throat. | Move the interesting sentence to the front. Nobody taps "more" for a warm-up. |
Do not gate anything on a detector score. In a 2023 study in Patterns, Liang and colleagues ran essays written entirely by humans through seven widely used GPT detectors. Work by non-native English writers was misclassified as AI-generated at an average false positive rate of 61.3 percent, and 19.8 percent were flagged by all seven at once. Native-speaker essays were classified almost perfectly.
Those detectors keyed on how predictable the text was, and second-language writers tend to use plainer vocabulary. The tools were not broken. They were penalising a legitimate human voice. On a 150 character bio they are worse than useless, because short text is exactly where they fail hardest.
The bio contains at least one detail a generator could not have invented about you.
Put the disclosure where it cannot be missed
If money, free product, an affiliate link or a family relationship sits behind a post, the FTC expects a disclosure that an ordinary person cannot avoid seeing. The 2023 revision to the Endorsement Guides is specific about what fails:
▪ Below the "more" cut. If a follower has to expand the caption to see it, it does not count.
▪ Inside a hashtag block. Do not park #ad at the end of your tags. The guidance says not to mix a disclosure into a group of hashtags.
▪ In your bio only. A profile page disclosure does not travel with the post, and nobody visits your profile first.
▪ Platform tools alone. A paid-partnership label may help, but the responsibility rests with you and the brand, not the app. Add your own words too.
▪ Vague words. #ambassador, #partner and #comped are not treated as sufficient. "Paid partnership with X" and "X sent me this for free" are.
Then there is the AI question. Google's position on AI-assisted content is that quality matters more than method, and that you should disclose when a reader would reasonably wonder how something was made. Nobody wonders who wrote your hashtags. The EU AI Act is narrower and firmer: from 2 August 2026, AI-generated text published to inform the public on matters of public interest must be disclosed, unless a named person genuinely reviewed it and takes editorial responsibility.
The two-minute log
For anything commercial, keep a record. Which tool wrote the draft, who reviewed it, what claims were checked, whether the post needed a disclosure, and the date. It is boring, it takes two minutes, and it is the only thing that helps you if a claim is ever questioned.
YOU ARE DONE WHEN
A follower could see the disclosure without tapping anything, and a named human is willing to defend every claim in the post.
Quality control does not end at publish
Everything above is pre-flight. The loop after it is where AI-written profile copy quietly rots: the limit changes, a tag gets restricted, the feature in your bio is discontinued, and your confident line becomes a confident lie.
| SIGNAL | WHAT IT USUALLY MEANS | YOUR MOVE |
|---|---|---|
| Profile visits up, follows flat | The content works. The bio does not close. | Rewrite the first line of the bio. It is the only line most people read. |
| Reach normal, saves and shares near zero | People saw it and felt nothing worth keeping | Move the hook into the first 125 characters and cut the rest |
| One post reaches far fewer than its siblings | Possibly a restricted tag, possibly just the algorithm | Check the tags against banned lists before you rewrite the content |
| Search appearances rising over months | Your labelling is working as labelling | Keep the tag set stable for that content category. Do not chase trends. |
| A follower asks whether a post was sponsored | Your disclosure was not unavoidable | Fix the post, then fix the template that produced it |
Put a recurring reminder in the calendar to re-check platform limits every quarter. In this niche, ninety days is roughly how long a spec stays true.
If you only have fifteen minutes
01 Paste the bio into the app, not into a character counter. The app is the only thing that decides.
02 Delete every claim you cannot produce a receipt for in ten seconds.
03 Cut the hashtag block to five and give each one a different job.
04 Check the tags against a banned and restricted list.
05 Say the handle out loud to someone and watch them type it.
06 Read the bio aloud. Delete anything you would not say to a person.
07 Move the hook into the first 125 characters of the caption.
08 Put the disclosure above the "more" cut, in plain words, outside the tags.
09 Log who reviewed it and set a date to check the specs again.
About fifteen minutes once it is habit. The first time will take an hour, mostly spent deleting hashtags.
The profile is a promise
I keep thinking about those twenty-eight hashtags. What bothered me was not that the tools were wrong. A model trained on the old internet giving old advice is about as surprising as a calculator overflowing. What bothered me is that the output looked so much like expertise that I almost pasted it.
That is the actual risk in this niche. Not that AI writes a bad bio, but that it writes an average one instantly, with the confidence of someone who has read every bio ever written and understood none of them. Tools like BioGPT can help shape a stronger first draft, but they still need a human check before anything goes live.
None of these checks are new. Good social managers have always read the caption aloud, checked the handle and known the current limits. What is new is that nothing else in the pipeline will do it for you. The generator will not flag its own staleness. Instagram will not explain why a post underperformed. The follower who spots the invented Forbes mention will simply not follow.
So use BioGPT to get past the blank page, then run the five checks. Put your name on it, because a profile is a promise: this is who I am, this is what I make, and everything on this page is true.