Image posts underperform on X, even with high impressions

Most B2B teams blame the format when image posts flatline on X. On accounts SocialNexis monitors, those posts pull strong impressions in the first 15-minute window and then the amplification step never fires. That reads as an image problem in the dashboard. It is a posting-behavior problem.

What the three big datasets measure about X image and text reach

Ordinal's 87,528-post dataset, covering January 2024 through May 2026, gives the cleanest read on distribution. Text posts averaged 46,731 impressions. Image posts averaged 13,053. That is a 3.6x gap in favor of text, and it holds across the account types in the sample. If you have ever watched a plain text post outrun a carefully built graphic published by the same account in the same week, that gap is the number behind the feeling.

Then the same dataset flips direction on the second metric. Image and GIF posts convert at a 0.99% engagement rate. Text posts convert at 0.67%. Images reach far fewer people and get a larger share of them to do something. Both numbers come from the same 87,528 posts, which is what makes this comparison useful: it is not two studies disagreeing, it is one study measuring two different stages of the same funnel.

Buffer's 45M-post study measures median engagement rate and lands somewhere else again. Text posts hit 3.56%, image posts 3.40%. Text wins by roughly 5%, and that is the narrowest margin of any platform in the study. X is the only major platform Buffer looked at where text beats images on that metric at all. The margin is thin enough that it would be easy to dismiss, but the direction is consistent with everything else on this page.

Adilo's 3,200-post study adds the part that explains the disagreement. Image posts show the lowest average engagement of any organic format at 2.09%, against 3.24% for text. But on the median, images come out ahead: 1.28% versus 1.13%. That combination has one explanation. Text produces extreme variance. Adilo's sample includes viral text outliers exceeding 400% engagement, and a handful of those drags the text average far above where the typical text post lands. Images produce boring, consistent, mid-tier results.

Averages describe text posts badly and image posts well. That single fact accounts for most of the apparent contradiction between these datasets. If you benchmark your B2B account's text performance against a text average, you are benchmarking against a distribution whose mean is set by content you are not producing. If you benchmark image performance against an image average, the average and the median are close enough that the comparison is roughly honest.

Text wins distribution reach, images win conversion rate per impression, and text's headline average is set by viral outliers most B2B accounts will never generate. None of that explains why one specific B2B account's image posts underperform. For that you have to look past the format entirely, which is the rest of this guide.

Do image posts on X underperform for B2B, or does the data depend on account type?

There is no documented format penalty against images in anything X has published. The opposite, if anything. X's 2023 open-sourced recommendation algorithm applied a 2x engagement-score multiplier to posts containing images or video during Heavy Ranker scoring. Read quickly, that looks like a platform-level thumb on the scale in favor of visual content, which is exactly how it has been reported for three years.

The detail that gets dropped is what the multiplier operates on. It is a weight applied to engagement score, not to impressions. In the same scoring structure, likes carry a 0.5x weight and retweets carry 20x. A retweet is worth an enormous amount relative to a like, and the media multiplier does nothing until someone has already engaged. A text post that gets retweeted will outscore an image post that gets liked, media bonus included. The multiplier amplifies engagement that already exists. It does not create distribution from a standing start.

That is the mechanism behind the Ordinal split. Image posts converting at 0.99% against text at 0.67% is consistent with a media bonus operating on engagement. Text posts pulling 46,731 impressions against 13,053 is consistent with that bonus doing nothing at the distribution stage. Both numbers fall out of the same scoring design.

There is a larger caveat on all of this. X rebuilt its algorithm in January 2026 on a Grok-powered transformer model from xAI, replacing the 2023 open-source code. Every specific claim about a 2x image multiplier traces back to that 2023 GitHub release, including the claim in the paragraph above. No official breakdown of the current system's format weights has been published. Treat the 2023 numbers as the last thing we know for certain rather than as current fact, and be suspicious of any 2026 guide that cites them without saying so.

The second problem with the public data is who is in it. Buffer, Adilo, and Ordinal all pool creators, media accounts, consumer brands, and B2B companies into one sample. A B2B software company posting a changelog and a creator account building a personal following are not behaviorally comparable. They post at different cadences, to different audience compositions, with different reply dynamics, and the pooled benchmarks treat them as one population.

So the honest answer to the section heading is: the ranker does not appear to penalize image format directly, and the studies cannot tell you whether B2B accounts specifically suffer more. What we can say from monitoring B2B accounts directly is that when image posts underperform badly, account tier, posting cadence, and image content type explain it more often than the format does.

What most X image vs text comparisons miss about B2B reach

The three largest public studies on X post format do not segment by B2B versus consumer, by posting method, or by Premium tier. Those three variables move format performance more than the format itself in most B2B scenarios, and none of them are controlled for in the benchmarks B2B teams are comparing themselves against.

The posting-method gap is the one we can speak to directly. Across SocialNexis-monitored B2B accounts, rate-limited text-only posting runs, meaning runs where cadence is governed by API rate limits rather than by human pacing, produce a recognizable signature. Impressions stay high. Reply rates drop to near zero. The post plateaus without the second-wave amplification that genuine text-heavy thought leadership content receives after its initial distribution.

That signature is dangerous precisely because it looks like success. An impressions dashboard shows a text post reaching a large audience, the number is comfortably above the account's image posts, and the obvious conclusion is that text is working and images are not. The engagement mechanics underneath are broken. Wide distribution with no replies is not a well-performing post, it is a post the algorithm served and the audience ignored, and repeated often enough it feeds back into how the account is scored.

Image posts from those same accounts behave differently. Impressions start lower from the outset, but engagement rate per impression comes in higher, which is exactly the relationship the Ordinal dataset describes at 0.99% against 0.67%. The suppression pattern is not equivalent across formats on the same account, which is why a single format ranking cannot diagnose anything. You have to look at the shape of each format's curve separately.

There is also a statistical population problem in the text numbers. Adilo's 3.24% text average against a 1.13% median is driven by viral outliers from high-follower accounts, some clearing 400% engagement. A B2B company account posting product updates to a modest follower base is not in that population, even though it is averaged into the same dataset. Benchmarking against the mean sets a target that the median account in the study does not hit either.

The result is that most B2B format decisions are being made against numbers that describe a different kind of account doing a different kind of posting. That does not make the numbers wrong. It makes them the wrong baseline for deciding whether your own image posts are broken.

Premium tier outweighs post format for X B2B reach

Before format, check the account tier, because the tier effect is larger than every format effect in this guide combined. Non-Premium accounts posting images on X effectively receive near-zero measurable engagement. Premium account image posts average around 0.90% engagement rate. Premium and Premium+ accounts see 6 to 15x more impressions than free accounts, and that multiplier holds regardless of post format.

A 6 to 15x impression range dwarfs a 3.6x format gap. In practice that means a Premium account posting images will consistently beat a non-Premium account posting text, no matter which direction the platform-wide medians run. If your B2B team is running an unsubscribed account and debugging why image posts underperform, the format question is not the one that changes your outcome.

Set this against the macro picture and it gets worse. X's platform-wide average engagement rate was 0.12% in 2025 according to Socialinsider's analysis of 70M posts, down from 0.15% in 2024. That is the real baseline for the platform as a whole. Now look at the format benchmarks: Buffer's 3.56% for text, Adilo's 3.24%, Ordinal's 0.67%. Every one of those describes a sample sitting well above the platform mean.

Those studies are measuring accounts that are already winning. Buffer's data comes through a scheduling product, Ordinal's through an analytics product, Adilo's from a creator-heavy sample. All three select for accounts engaged enough with their own performance to be running measurement tooling. That is a filtered population, and a B2B account comparing its own sub-1% engagement rate to a 3.56% published median is comparing itself to a sample it was never in.

The sequencing that follows is unglamorous. Resolve tier first, then posting behavior, then format mix. For B2B teams evaluating whether X deserves budget at all, running the test without Premium produces a result that tells you nothing about the channel, only about the tier. We have seen teams write off X on data collected entirely from free accounts, which is a decision made on a variable they did not know they were testing.

Threads, not images: the B2B X format with the highest engagement rate

The highest-engagement organic format on X for B2B is not text and not images. Ordinal's 87,528-post dataset puts threads at a 1.89% engagement rate, nearly 3x text-only posts at 0.67% and roughly 3.5x video at 0.53%. Nothing else in the format comparison comes close, and threads are missing from almost every format guide on this topic.

The omission is structural. Most format studies benchmark text, image, and video because those are the three categories a posting API reports cleanly. A thread is several posts with a parent-child relationship, which takes deliberate work to classify, so it gets dropped or collapsed into the text bucket. That means the format with the best engagement rate on the platform is systematically underrepresented in the data teams use to pick formats.

For B2B specifically, threads solve a problem the other formats do not. They build engagement without depending on visual assets and without depending on the impression-distribution advantage that Premium tier and a large following provide. An account that cannot buy its way to distribution can still generate interaction density, and interaction density is what the scoring system rewards most heavily. Recall the weight structure from the 2023 code: retweets at 20x against likes at 0.5x. Engagement depth beats engagement volume.

Threads also create more surface area for replies. Each post in a thread is a separate object that can be quoted, replied to, or bookmarked on its own, and a multi-post breakdown of a B2B process typically generates more reply activity than a single image post carrying the same information. That matters more than it sounds, because reply activity is the signal most closely tied to the second-wave amplification that image posts on automated accounts never get.

The format also matches how B2B buyers read X. Multi-part breakdowns of a market situation, a deal structure, a pricing decision, or a category comparison hold attention across posts. A single image containing the same content gets one impression event and fades. If your B2B goal on X is authority rather than impression volume, threads are the format to build the calendar around, and the 1.89% figure is the strongest argument in this guide for changing what you publish rather than how you publish it.

How posting cadence and image-to-text ratio change X image post reach

The most-cited evidence that scheduling tools do not hurt reach is Buffer's controlled test from November to December 2017: third-party scheduled posts averaged 21,108 impressions against 21,671 for native posts, a gap of about 2.6% that Buffer described as statistically negligible. Link posts scheduled through third-party tools even outperformed native ones in that test. The finding is real and it is also nine years old. It predates X's current automation enforcement architecture entirely, and it should not be used to reason about API posting behavior today.

What X does document is thin. The company now officially calls shadowbanning "visibility filtering," and the stated triggers are multiple user reports, flagged keywords, aggressive tagging, and rapid replies. There is no documented "posted via third-party app" label anywhere in the ranking system. That is worth stating plainly, because a lot of automation advice assumes one exists. High-cadence automated posting is a behavioral signal, and behavioral signals are evaluated separately from content type.

On B2B accounts SocialNexis monitors that run scheduled image posts through third-party APIs at cadences above 4 to 6 posts per day, we see a distinct pattern that does not match the organic reach loss text-only posting shows at equivalent cadence. The image posts get their early impressions in the first 15-minute window at normal levels. Then the amplification step, where strong early engagement normally triggers wider distribution, is truncated. The curve just stops climbing.

In an analytics view, that looks identical to a format problem. The post has fewer total impressions than the account's text posts, images are the obvious variable, and the conclusion writes itself. The truncation is a posting-behavior signal, not a content-type signal. The same account posting the same image at a lower cadence does not show the same cutoff.

The ratio effect is the more useful finding. Across monitored accounts, the image-to-text ratio inside a rolling 7-day posting window correlates with distribution outcomes differently for API-posted content than for natively-posted content. Accounts posting more than 70% images through automation consistently hit lower per-post impression ceilings than accounts alternating text and image at 50/50 or 40/60. The images themselves are not the variable that changed. The sameness of the posting pattern is.

The most plausible reading is that format homogeneity across a window functions as an inauthenticity heuristic. A human running a B2B account produces a messy format mix because their week is messy: a quick reaction post, a screenshot, a thread, a link, a graphic someone in marketing finished on Thursday. A queue loaded with fourteen graphics does not look like that. Nothing in this requires the platform to dislike images, and everything in it is fixable by varying what goes into the queue.

SocialNexis data: suppression patterns in B2B accounts posting image-heavy automated content

There is a second failure mode inside the image category itself, and it is the one that catches B2B marketers hardest. When a monitored B2B account posts an image containing dense text, a stats graphic, a quote card, a slide screenshot, we observe lower reach than clean-image posts on the same account, controlling for posting time and account status. Same account, same schedule, same tier, different visual content type, different outcome.

The likely mechanism is image content analysis in the ranking pipeline, which the open-source code context references, scoring text-heavy images differently from photographic or original visual content. We cannot see inside the model to confirm it, and after the January 2026 rebuild nobody outside xAI can. But the pattern is consistent enough across accounts that we plan around it.

This lands directly on how most B2B teams source images. Repurposed slide decks, infographic exports, and quote cards are the default image library at a B2B company, because they already exist and someone has already approved them. That entire library is the worst-performing image type we measure. The format is not the problem. The visual content type is, and a team that switches from graphics to text posts to fix reach has fixed the wrong variable.

Links complicate the picture further. X's external link penalty, long estimated at a 30 to 50% reach reduction, was officially confirmed removed by Elon Musk in July 2026. Suppression reportedly continues through indirect mechanisms including delayed redirects and engagement-based algorithmic demotion. A B2B account attaching an external link to an image post is not operating in a clean environment even after the official reversal, and treating the announcement as a full all-clear will overstate what you should expect from link posts.

The most revealing thing we saw was in the recovery, not the penalty. Monitoring B2B accounts through the weeks following that July 2026 reversal showed manual posters getting link post reach improvements within 48 to 72 hours of the announcement. API-posting accounts sharing the same domains saw improvement lag by 1 to 2 weeks. Same links, same destination sites, same window, different recovery speed.

A purely algorithmic reach restoration would not behave that way. A rule change applies to everyone at once. A staggered recovery that sorts accounts by posting method points to a trust-score component that favors human-consistent posting patterns, with automated accounts re-earning distribution rather than being handed it back. Practically, that means the accounts most likely to use scheduling infrastructure, which is to say most B2B teams, got the smallest and slowest benefit from the policy change that was supposed to help them.

Build a post format mix that works within X's automation detection signals

Start with the ratio, because it is the single change with the clearest measured effect. Keep your rolling 7-day image-to-text ratio at or below 50/50 when posting through automation tools. Monitored accounts exceeding 70% image posting via API hit measurably lower impression ceilings than accounts running 50/50 or 40/60. This is not a stylistic preference about variety in the feed. It is a behavioral signal, and the queue is where you control it.

If you are linking out, attach the image yourself. Agorapulse's controlled experiment found native image uploads averaging 92.67 impressions against 81.80 for Twitter Card link previews, roughly a 13% advantage. The card preview is the path of least resistance because X generates it from the URL automatically, which is precisely why it is worth overriding. Upload the image to the post and put the link in the copy.

Swap graphics for threads on your authority content. Threads run at a 1.89% engagement rate, nearly 3x text and 3.5x video in Ordinal's data, and they do not depend on the visual assets that are causing half the problems in this guide. For a B2B team, the practical mix is threads as anchor content, text posts for distribution volume, and images held under half of weekly post volume on automated accounts.

Audit your image library against the dense-text finding. If your image posts are predominantly slide exports, stats cards, and quote graphics, you are publishing the exact visual type that reaches least on the accounts we monitor. That does not mean abandoning data visuals. It means not making them the entire image strategy, and it means testing a clean-image post against a text-heavy graphic on your own account before assuming the finding transfers.

One comparison settles the format question. If your image posts show high impressions and near-zero replies and retweets, check whether the same plateau appears on your text posts. If both formats flatten the same way, the problem is posting behavior: cadence, ratio, or automation trust, and switching formats will change nothing. If only the images stall while text posts pull normal reply activity, then you are looking at format-level or visual-content-level signals and the fixes above apply.

X's platform-wide average engagement rate was 0.12% in 2025, down from 0.15% in 2024. Everything in this guide operates inside that number. Format mix, ratio discipline, and thread structure move a B2B account meaningfully within that range, and none of them make X behave like a channel it is not. Decide what you want from the platform before optimizing the format, because the format is the smallest of the variables that determine whether the answer is worth the work.

Frequently asked questions

Why do my image posts on X get fewer impressions than my text posts even though images supposedly boost engagement?

The impression gap is real: text posts average 46,731 impressions vs 13,053 for images across 87,000 posts (Ordinal, 2024 to 2026). Images convert at a higher engagement rate (0.99% vs 0.67%) but distribute to fewer people. If your image posts specifically stall after the first 15 minutes with no second-wave amplification, that pattern points to posting-cadence suppression on automated accounts rather than a format-level penalty from the algorithm.

Does the X algorithm treat images differently depending on whether a post was scheduled via a third-party tool vs. posted natively?

X does not officially document a 'posted via third-party app' penalty in its algorithm. Buffer's 2017 controlled test found only a 2.6% impression gap between third-party and native posting. That test predates X's current automation enforcement by six years. What SocialNexis observes now is that high-cadence automated accounts see a truncated amplification window regardless of format, suggesting the signal is posting frequency and behavioral pattern, not the API source alone.

Do image posts on X underperform for B2B accounts specifically, or is the text-beats-image finding true across all account types?

The major studies (Buffer, Ordinal, Adilo) pool all account types without B2B segmentation. The text-beats-image finding in impression counts is consistent across datasets, but the degree of underperformance is likely worse for B2B accounts using automation at high cadence. Non-Premium B2B accounts see near-zero engagement regardless of format, making the format comparison largely secondary until the account tier question is resolved.

If I add an image to a post that also contains an external link, does the image offset the link suppression?

X's external link penalty was officially confirmed removed in July 2026, but indirect suppression through delayed redirects and engagement-based demotion reportedly continues. Agorapulse's experiment shows native image uploads outperform Twitter Card link previews by about 13% in impressions, so attaching an image separately is a marginal improvement. It does not appear to fully offset link-related distribution losses on accounts that post through automation tools.

Why does my image post have high impressions but almost no likes, replies, or retweets?

High impressions with near-zero interaction is a distinct engagement signature SocialNexis associates with rate-limited posting behavior or automation detection, not format underperformance. When the algorithm serves a post widely but those users do not interact, it suggests the account's behavioral signals have flagged it as lower-trust. The diagnostic check: if the same plateau appears on your text posts, format is not the variable. The problem is in the posting pattern.

Do native image uploads on X outperform Twitter Card link previews for B2B posts?

Yes. Agorapulse's controlled experiment found native image uploads average 92.67 impressions versus 81.80 for Twitter Card link previews, a roughly 13% advantage. For B2B posts that need to include an external link, the better structure is to attach an image directly to the post and include the link in the copy text, rather than letting X generate the card preview automatically from the URL.

How does posting cadence or inter-post timing affect image post reach on X for B2B accounts using automation?

SocialNexis data shows that B2B accounts posting more than 70% images via automation in a rolling 7-day window consistently see lower per-post impression ceilings compared to accounts alternating text and image at 50/50 or 40/60 ratios. Format homogeneity in a posting window appears to function as an inauthentic-posting signal. Reducing the image-to-text ratio addresses this more reliably than switching from images to text alone.

What is the difference between X visibility filtering and organic reach loss from image posts?

X's 'visibility filtering' (their official term for shadowbanning) is triggered by user reports, flagged keywords, aggressive tagging, and rapid replies. It suppresses a post across multiple platform surfaces. Organic image reach loss is different: it shows up as a lower initial impression count compared to text posts on the same account, without platform-wide suppression. You can distinguish them by checking whether your text posts reach normally while only images underperform, which points to format-level signals rather than account-level filtering.

Should B2B brands on X prioritize threads, text-only posts, or image posts for authority content?

Ordinal's 87K-post dataset puts threads at 1.89% engagement rate, nearly 3x text (0.67%) and 3.5x video (0.53%). For B2B authority content, threads are the highest-engagement organic format on X. Text-only posts distribute more widely but with high variance. Image posts distribute least widely but convert at the highest rate per impression. A practical mix: threads for anchor content, text for distribution volume, images kept under 50% of weekly post volume on automated accounts.

Does the January 2026 Grok-powered X algorithm update change how image posts are ranked?

Potentially. X's algorithm was rebuilt in January 2026 using a Grok-powered transformer model from xAI, replacing the 2023 open-source Heavy Ranker code. The 2x image engagement-score multiplier documented in the 2023 GitHub release may not reflect how the current system scores format. No official breakdown of the Grok-based system's format weights has been published, so format comparisons citing the 2023 code should be treated as dated guidance rather than current fact.

Sources and further reading

• Buffer's 45M-post content format study comparing text vs. image performance on X
• Ordinal's 87,000-post Twitter analytics breakdown with impression and engagement rate data by format (2024 to 2026)
• X Engineering's 2023 recommendation algorithm documentation