Who Gets to Hear One Another? Gatekeeping in Indian Country

My deepest concern is not whether a single video becomes popular. It is whether we can hear one another—and who bears the costs when the systems we rely on get in the way.

Sometimes someone brings us a concern that, at first, seems to belong to a particular field, community, or line of work. They have been paying attention to something and want us to take a closer look. When we do, we may find that the consequences reach many more people than the original story suggests.

That is how this essay began. A relative sent me an Instagram post he wanted me to see. In the account I received, Andrew Callaghan shared his concern that Channel 5’s Native-focused reporting was being shadowbanned. I like Channel 5. I appreciate the gritty, on-the-ground reporting and the encounters with people and places I rarely see covered elsewhere. I wanted to take the concern seriously, without treating my appreciation for the reporting as proof that its explanation was correct.

But the more I considered the issue, the less it seemed to be a question about Channel 5 alone—as a company, a journalism project, or a channel trying to reach its subscribers. I do not have the evidence needed to settle what happened to that particular documentary. My concern is the larger question: what happens to our ability to understand and respond to one another when communication passes through systems whose decisions we cannot easily see or challenge?

In Indian Country, that question is not only about whether the wider public hears Native voices. It is also about whether we hear one another: whether a concern reaches someone who can recognize it, whether useful knowledge finds someone who needs it, and whether people working on related problems discover that they are not working alone. These are purposes of communication that cannot be reduced to making a post popular.

“Gatekeeping” here means selecting, ranking, restricting, or otherwise organizing the paths by which communication reaches people. It describes a function to investigate, not a conclusion that every selection is wrongful. A service can make communication possible while also setting consequential conditions on it.

More than a matter of reach

Tarleton Gillespie’s Custodians of the Internet examines how platforms organize not only access to material but relationships between speakers and audiences. It also describes why leaving a platform can be costly: an alternative service does not automatically carry over a person’s relationships, accumulated interactions, or professional connections. These are features of his historical analysis, not findings about the documentary. ¹

Indigenous scholarship gives those relationships a more specific context. In Indigenous Digital Life, Bronwyn Carlson and Ryan Frazer examine community connections, care, identity, creativity, and future-oriented activity. Their account principally concerns Indigenous Australians, approximately 2010–2020. It shows why digital participation should not be reduced to either audience totals or a story of passive exclusion. ²

A more directly relevant account of research in Indian Country appears in Marisa Elena Duarte and Morgan Vigil-Hayes’s chapter “How We Connect.” Their Indigenous feminist approach treats relationships and contextual knowledge as central to digital research. People knowledgeable about Native communities help determine what researchers should look for, rather than appearing only as data to be collected. ³

The implication for this inquiry is practical. A platform-wide measure of popularity cannot, by itself, answer whether people can maintain a connection they value. A class announcement reaching thirty people who need it may accomplish its purpose better than a broadly popular post reaching thousands who cannot use it. This is an example of how purpose changes evaluation, not a proposed universal performance threshold.

The 2016–2018 Reclaiming Native Truth research supplies context for the separate problem of public representation: familiar historical images could coexist with limited awareness of contemporary Native life. Nielsen’s 2021 television report also identifies Native underrepresentation within its own analysis. Neither study measures communication between Native users on a current platform. Public visibility and the ability to reach one another must not become substitutes for each other. ^{4 5}

The costs of fragmented understanding

When gatekeeping materially obstructs a connection, the cost need not remain with the person whose post received less attention. A community organization might spend scarce money repeating an announcement. A family might miss an opportunity. People addressing the same problem might duplicate work because neither knows what the other has already learned. An unanswered appeal might be mistaken for indifference when it never reached the people who would have responded. These are possible consequences to examine, not outcomes established by the Channel 5 account.

This is what concerns me about fragmented understanding. I do not mean that everyone should encounter the same information or arrive at the same judgment. Different interests, disagreement, and privacy are part of a healthy communicative life. The danger is that we may form judgments about one another from incomplete accounts without knowing what is missing—or why it is missing.

The language of externalities can make the costs sound distant. But the people bearing them may have to spend more money, repeat unpaid work, recover a missed opportunity, or repair a relationship strained by misunderstanding. A cost that does not appear in a platform’s measures of success does not thereby disappear from someone else’s life.

Not every communication failure comes from a platform decision. People may be unavailable, uninterested, unable to connect, or using different settings. Naming possible costs gives an investigation something to look for; it does not establish what caused them. But when a consequential barrier is identified, the response should consider the people who bear its effects, not just whether a technical setting has been corrected.

What a feed serves—and what people need

A system can help people discover unfamiliar material, organize a crowded feed, or reduce harmful exposure. Those functions may be useful. They do not answer every question about whether the system supports an existing relationship.

Consider two requests: “Show me something I might enjoy” and “Help me keep up with this community organization.” A service may use recommendations in both situations, but the expectations are different. An inference about likely engagement should not be treated as identical to an express choice to follow a source. The ethical argument concerns how that difference is represented, supported, and made controllable.

This is not a demand to remove selection or moderation. Spam, harassment, and unwanted exposure can themselves make communication difficult or unsafe. Gillespie’s account emphasizes the demands and competing purposes of moderation, rather than imagining a viable platform with no rules. The question is how those purposes are pursued and what happens when the system makes consequential mistakes. ¹

What the research can—and cannot—show

Availability is not the same as delivery

Gillespie documents ways platforms have retained content while limiting its discoverability or circulation. His later article, “Do Not Recommend?”, examines reduction through recommendation systems. The fact that a video remains online therefore cannot, on its own, establish that its circulation is unrestricted. Equally, a disappointing number does not establish that a restriction occurred. ^{1 6}

The route matters. YouTube describes Home as personalized. Its notification guidance distinguishes personalized notifications from other settings and points subscribers to a separate Subscriptions feed. Its analytics distinguish impressions, views, unique viewers, and traffic sources; “Browse features” includes several surfaces. These distinctions are the company’s published account, not an independent test of any particular upload. ⁷

That gives a complaint a more useful shape: was material unavailable, ineligible for recommendation, ranked lower, missing from a selected feed, or not notified? Was it encountered but not watched? Different observations require different explanations. A following route may exist without guaranteeing that every person will see every post.

People’s observations count, even when the mechanism is hidden

Kelley Cotter’s research on the Instagram shadow-banning dispute examines the authority to say what an algorithm is doing. Her qualitative study draws on seventeen influencer interviews and a wider record of online discussion and platform statements. It does not estimate how often shadow banning occurs. ⁸

Cotter argues that users have situated knowledge that official accounts do not exhaust. She also cautions that access to technical details does not give platform owners complete knowledge of every consequence. Her term black box gaslighting belongs to her analysis of that dispute; this inquiry does not apply it to YouTube or presume knowledge of anyone’s intentions. ⁸

The distinction is between taking an observation seriously and accepting its proposed cause. A community should not have to obtain inaccessible internal records before a well-specified complaint can be heard. At the same time, relevant testimony does not automatically settle mechanism, scale, or intention.

Engagement partly reflects the opportunities a system provides

Allison J. B. Chaney, Brandon M. Stewart, and Barbara E. Engelhardt study algorithmic confounding: systems learn from behavior that earlier recommendations have already influenced. In their simulations, repeated training can increase overlap in what users consume beyond what the modeled ideal requires, without corresponding gains in utility. ⁹

The feedback sequence is straightforward: recommendations influence encounters; encounters influence choices; those choices become training data; later recommendations reflect those data. This is not evidence that the sequence caused the documentary’s reported shortfall. It explains why observed engagement should not automatically be treated as an independent measure of what people would value having an opportunity to encounter.

The paper studies simulated behavior, not the convergence of real people’s underlying preferences. Its discussion of poorly represented preferences is not a finding about an ethnic population. Its “utility” is a modeling quantity, not a measure of a language’s or community’s worth. These distinctions prevent a relevant mechanism from becoming an invented Native-specific result.

Duarte and Vigil-Hayes add a related caution: data are traces of social reality, not that reality itself. Some people do not generate the available traces, and the conditions of data collection affect what can be learned from them. Their argument reinforces the need to interpret metrics with community knowledge rather than replacing that knowledge with metrics. ³

Financial pressure is relevant, but it is not proof of a scheme

Patreon’s 2025 State of Create reports that 78 percent of surveyed creators said algorithms affected what they created, while 56 percent said algorithms prevented them from exploring passions and interests. These are self-reports from a U.S. survey conducted with NewtonX in August 2024, not verified restriction events or a Native-specific result. Patreon promotes its own commercial alternative and has an interest in this subject. ¹⁰

An inquiry into costs should distinguish lost revenue, additional promotion spending, and unpaid work from an allegation that a platform intentionally reduced reach to sell it back. The first set can be investigated without presuming the second. Buying promotion also does not, by itself, reveal what caused an earlier organic shortfall.

Advertising eligibility, paid promotion, and ordinary distribution should not be collapsed. YouTube, for example, states that monetization status itself is not a recommendation signal, while the content of a video can affect both advertising suitability and recommendations. That is an actor statement relevant to the questions asked, not an independent guarantee about implementation. ⁷

No loss amount, improper payment, or right to compensation is established here. The point is that costs deserve examination rather than disappearing behind an aggregate engagement figure.

What makes gatekeeping answerable?

The argument has three steps. People’s ability to sustain chosen relationships and pursue worthwhile activities matters. A platform that controls a consequential part of those connections can influence the conditions under which people succeed or fail. When that influence creates a credible risk of substantial interference, affected people need a practical way to understand the relevant conditions, question a mistake, and seek a proportionate response.

This is an ethical argument, not a legal ruling or a finding that every missing post creates a duty to compensate. The appropriate response depends on the service’s role, reasonable expectations, severity, available evidence, alternatives, and the costs of investigation. It should protect recipients’ choices and safety as well as speakers’ interests.

A platform does not owe every speaker maximum exposure. But maximum exposure is not what someone requests merely by asking whether a communication they chose to receive has been restricted. An adequate account of the system should help people distinguish the promise of a service from what they assumed it promised.

The Santa Clara Principles offer an established reference point. Their second iteration treats enforcement-related algorithmic downranking as an action deserving transparency, alongside removal and account suspension. They call for understandable rules, notice, appeal, and attention to language and cultural context. They are civil-society recommendations, not a statute or evidence that any company has complied with them. ¹¹

Their enforcement focus should not be confused with a demand for an explanation of every recommendation. I would distinguish three practical needs.

Know what happened when a restriction was imposed. A person should be able to learn what was affected, the relevant reason, the scope and duration, and how to contest a mistake. Where disclosure must be limited to protect another person or prevent abuse, the remaining explanation should still support meaningful review. “The system did it” is not enough to distinguish a justified restriction from an error.

Have usable choices about ongoing communication. People should understand which routes prioritize discovery and which support sources they have chosen. Where ordinary ranking—not a discrete enforcement decision—is the concern, accessible controls, clear service descriptions, and investigation of documented recurring discrepancies are more useful than pretending every missed impression can receive an individual adjudication.

Address the consequences of a verified error. Correcting a classification matters, but it may not restore a missed registration window, a lost day's work, or the labor of rebuilding contact. A response should examine what remains, who can act on it, and what repair is feasible. Financial redress would require its own evidence of causation, responsibility, and loss; it should neither be assumed nor ruled out merely because the error was automated.

These are proposed requirements for answerability, not a claim that the cited research measured their implementation. They go beyond making information visible: the information must help someone act, challenge, or obtain an explanation of a genuine limit.

Understanding a problem without taking over a relationship

A useful inquiry could begin with a specific communication, the intended audience or relationship, the route selected, the observed discrepancy, and the consequence. It could compare available records over equivalent periods while keeping the difference between exposure and response visible. Creator observations, recipient accounts, platform records, and contextual knowledge can contribute different parts of the explanation.

No one should have to disclose a community’s entire contact network or prove Native identity to receive an explanation about their account. Research into group-level effects should use appropriate consent, protected data, and community-informed categories. The existence of a question about Native communication is not permission to infer people's identities or collect their relationships indiscriminately.

Nor is Indian Country a single audience with one preference. People may want different connections, disagree with community leaders, leave a group, mute a speaker, or keep some knowledge within a limited circle. Making communication possible must not become a demand that everybody hear the same voices. Carlson and Frazer’s attention to diverse forms of community and the ethics of using apparently public material is relevant to that boundary. ²

What remains open

The reviewed sources establish relevant distinctions and research findings about organized visibility, situated knowledge, recommendation feedback, and Indigenous digital relationships. They do not establish an internet-wide Native suppression rate, authenticate the reported one-fifth measure, or identify a corporate cause for the documentary’s distribution.

The ethical argument does not depend on settling those incident findings. It asks what people should be able to know and challenge about consequential control over their communication. Evidence that a specific failure came from a viewer's setting, a technical interruption, ordinary competition for attention, or an applicable restriction would change that incident's explanation. It would not make the broader question disappear; it would tell us which part of the system needs explanation or response.

The argument itself is also open to challenge. A proposed safeguard should be revised where it fails to help affected people, demands unavailable precision, exposes private information, undermines legitimate safety measures, or costs more than its expected benefit warrants. Transparency has to be useful and proportionate, not merely exhaustive.

Hearing one another

A relative’s invitation to look at one post brought me to a question larger than that post. Callaghan’s reflections can be heard as an invitation to investigate without turning every part of his explanation into an established fact. The transcript identifies the concern that prompted this essay; it does not settle the particular allegation.

The issue is not a guaranteed audience. It is whether people can understand and challenge the conditions under which they reach one another—and whether consequential mistakes receive a response that attends to the costs people bear.

Indian Country’s ability to hear itself should not depend on first proving that somebody intended to keep it silent. Investigating a consequential barrier and attributing a hidden motive are different tasks. Keeping them separate makes room for evidence, responsibility, and repair without treating either a creator’s suspicion or a platform’s account as the last word.

Sources and reading

1. Gillespie, Tarleton. 2018. Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That Shape Social Media. Yale University Press. Selected passages, especially pp. 5–12, 22–23, and 177–182. Publisher record.

2. Carlson, Bronwyn, and Ryan Frazer. 2021. Indigenous Digital Life: The Practice and Politics of Being Indigenous on Social Media. Palgrave Macmillan. Selected passages, especially pp. 4–13 and 247–248. Principally Indigenous Australia, circa 2010–2020. DOI.

3. Duarte, Marisa Elena, and Morgan Vigil-Hayes. 2021. “How We Connect: An Indigenous Feminist Approach to Digital Methods.” In Bronwyn Carlson and Jeff Berglund, eds., Indigenous Peoples Rise Up: The Global Ascendency of Social Media Activism, pp. 93–111. Rutgers University Press. Selected passages, especially pp. 93–97, 101–103, and 108. Collection.

4. First Nations Development Institute and Echo Hawk Consulting. 2018. Reclaiming Native Truth: Research Findings—Compilation of All Research. Findings report, especially printed pp. 8 and 18; research overview. One 2016–2018 research program, not two independent studies.

5. Nielsen. 2021. Official overview of Being Seen on Screen. Historical television representation context, not a measure of social-platform distribution.

6. Gillespie, Tarleton. 2022. “Do Not Recommend? Reduction as a Form of Content Moderation.” Social Media + Society 8(3). DOI. Analysis of policies and communications, not an audit of the incident discussed here.

7. YouTube Help. Performance FAQ, notification guidance, and video-reach definitions. Company descriptions consulted in the underlying research on September 26, 2026, not independent tests of the reported incident.

8. Cotter, Kelley. 2021, online publication. “‘Shadowbanning is not a thing’: black box gaslighting and the power to independently know and credibly critique algorithms.” Information, Communication & Society. DOI. References use the supplied online-first article’s pagination, especially pp. 5–6, 11, and 14–15. A qualitative Instagram study, not a prevalence estimate.

9. Chaney, Allison J. B., Brandon M. Stewart, and Barbara E. Engelhardt. 2018. “How Algorithmic Confounding in Recommendation Systems Increases Homogeneity and Decreases Utility.” Proceedings of the 12th ACM Conference on Recommender Systems, pp. 224–232. DOI. Simulations, not a field study of Native users.

10. Patreon. 2025. State of Create. Selected findings and methods. August 2024 U.S. survey with NewtonX, including 1,007 creators and 2,002 fans; industry-sponsored self-report evidence.

11. Santa Clara Principles coalition. 2021, second iteration. The Santa Clara Principles on Transparency and Accountability in Content Moderation. Chapeau, foundational principles, and operational notice/appeal principles. Civil-society recommendations, not enacted law or proof of implementation.

A note on the starting account

The account discussed here is represented in the research record by the supplied transcript, identified by its provider as Andrew Callaghan’s comments. The recording and channel analytics were not independently reviewed. The essay does not establish the specific shadow-banning allegation, its scale, or a financial loss.

This essay grew out of a Sift Room inquiry. Its source notes, claims ledger, and earlier versions remain in the research record. AI assisted source review, organization, drafting, and internal checks. Those checks are not independent validation. The ethical argument and illustrative examples are distinguished from the findings of the cited studies.