Blog · Review Essay · Last reviewed June 25, 2026

LikeWar and the Social Media Battlespace

P. W. Singer and Emerson T. Brooking's LikeWar is not mainly a book about misinformation. It is a book about conflict after the feed becomes an operating environment: war, politics, celebrity, propaganda, surveillance, and crowd behavior all passing through platforms that convert attention into visible force. In the AI era, its value is that it explains why synthetic media and persuasion tools enter a world already trained to treat virality as evidence.

For this review, the social media battlespace is the system in which ranking, recommendation, social proof, identity performance, open-source evidence, paid reach, coordinated behavior, and institutional reaction become part of the conflict itself. The governing question is not only whether a post is true, but whether a platformed path can turn attention into pressure before verification, law, or public memory catches up.

The operational unit is the attention-to-action chain: artifact, source, platform routing, amplification, institutional uptake, correction, and archive. LikeWar stays useful when it is read as a demand to map that chain rather than as a slogan that turns every online argument into war.

The Book

LikeWar: The Weaponization of Social Media was published by Houghton Mifflin Harcourt in 2018. National Defense University Press's review record lists the hardcover at 416 pages with ISBN 978-1328695741. Google Books lists the same 2018 Houghton Mifflin Harcourt edition with ISBN 1328695743 / 9781328695741 and a 405-page length field. HarperCollins, which now hosts the Mariner Books listing, presents the 2019 paperback as a book about the collision of war, politics, and social media.

The author pairing matters. P. W. Singer's New America profile identifies him as a strategist and senior fellow whose nonfiction includes LikeWar, Cybersecurity and Cyberwar, Wired for War, Corporate Warriors, and Children at War. Emerson T. Brooking's Atlantic Council profile identifies him as a resident senior fellow at the Digital Forensic Research Lab, focused on foreign information manipulation and information resilience. The book is written from the national-security and conflict-analysis world, but its implications are broader than military affairs.

It belongs beside The Chaos Machine, Invisible Rulers, Network Propaganda, The Filter Bubble, The Attention Merchants, Twitter and Tear Gas, and The Revolt of the Public. Those books explain platform incentives, networked crowds, asymmetric propaganda, attention capture, and institutional crisis. LikeWar adds the conflict layer: once public attention becomes a contested surface, persuasion, visibility, humiliation, recruitment, and operational deception all become strategic acts.

Current Context

As of June 25, 2026, LikeWar reads less like a forecast and more like an early map of the regulated information environment. The EU Digital Services Act now treats very large online platforms and search engines as systemic-risk systems: the Commission says the strongest DSA obligations apply to services with more than 45 million monthly users in the EU, and its designated-service list was updated on May 28, 2026. The DSA's Article 40 researcher-data-access regime also moved from principle toward procedure when the Commission adopted a delegated act on July 2, 2025. That is directly relevant to the book because the evidence of platform conflict lives in reach, ranking, ad delivery, enforcement queues, recommendation paths, and account networks, not only in individual posts.

The synthetic-media layer is also now formal. The European Commission says the AI Act's Article 50 transparency obligations for marking, detection, and labelling of AI-generated and manipulated content apply from August 2, 2026, and its June 2026 Code of Practice on Transparency of AI-Generated Content is meant to support compliance. NIST's Generative AI Profile treats provenance harms such as misinformation, disinformation, deepfakes, nonconsensual intimate imagery, and tampered content as risk-management objects; NIST AI 100-4 surveys provenance, watermarking, detection, testing, auditing, and maintenance as technical approaches to synthetic-content transparency. C2PA's 2.4 specification, published in April 2026, extends content-credential machinery with new assertions, asset-format support, and JSON serialization. Those controls matter, but they govern the artifact more easily than the campaign.

U.S. consumer-protection law has also touched the social-proof layer. The FTC's August 2024 final rule on fake reviews and testimonials addresses fake reviews, AI-generated fake reviews, and fake social media indicators in commercial contexts. That is narrower than LikeWar's political and military frame, but the principle travels: counters, testimonials, followers, likes, and shares become dangerous when an institution treats them as evidence of public reality without knowing how they were produced.

Election administration gives the clearest public-safety version of the problem. The U.S. Election Assistance Commission warns that AI tools can accelerate false or biased election information, make phishing and social engineering more effective, imitate official sources, and produce plausible but inaccurate voting information. Its practical response is not only content detection; it is official channels, voter-facing materials, verified sources, and prepared communication routines. That is the book's lesson in administrative form: trusted paths need to exist before the feed turns a false claim into pressure.

Foreign information manipulation has become a named governance category. The European External Action Service's FIMI reports map the digital infrastructure used by foreign actors to manipulate information spaces and describe cross-platform tactics rather than single-message falsehood alone. This is where the book's battlespace concept remains useful: the operation is the networked path by which a claim is created, amplified, legitimized, contested, archived, and sometimes acted on.

The practical update is that platform conflict now has a records problem. API access can close, private groups can hide distribution, screenshots can detach claims from origin, and AI systems can summarize or translate claims without preserving the source route. The public may see the artifact and the reaction while losing the path between them. That is exactly where governance has to insist on provenance, ad libraries, recommender evidence, researcher access, and incident records.

The 2026 update is therefore evidentiary, not theatrical. The useful question is not whether the internet has made everyone a combatant. It is whether the institution can distinguish lawful dissent, crisis testimony, ordinary error, paid influence, automated distribution, foreign manipulation, and coercive targeting before it reacts. The same public feed can contain all of those at once, and a safety regime that cannot tell them apart will either underreact to operations or overreact to speech.

The Feed as Battlespace

The strongest idea in LikeWar is that social media is not merely a channel through which conflict is discussed. It becomes part of the conflict. A platform makes speech measurable, ranks it, shows counters, connects strangers, rewards timing, amplifies emotion, and makes spectators visible to one another. That changes what power can do.

Old propaganda moved through newspapers, radio, posters, television, sermons, schools, and rumors. Platform propaganda inherits that history but adds live feedback. A message can be tested in public. A rumor can be iterated. A raid, atrocity, denial, meme, leak, and counter-narrative can compete in the same attention stream. Supporters are not only an audience; they become distributors, analysts, harassers, fundraisers, scouts, translators, mememakers, and pressure groups.

This is why the book is useful beyond the cases it recounts. It describes a change in the physics of public life. A government, militia, activist network, celebrity, extremist movement, newsroom, platform company, or private citizen can act inside the same visible environment, but they do not enter it with equal resources or equal risk. The platform makes them comparable as accounts while the real world keeps them unequal as institutions.

The result is not a clean replacement of tanks by tweets or territory by hashtags. It is an added layer. Physical events produce media; media changes perception; perception changes behavior; behavior changes the next event. A battle, protest, election, terror attack, police shooting, border incident, court case, or product launch now has a networked afterlife that may matter as much as the event itself.

The useful distinction is between speech, campaign, and operation. A post may be lawful speech. A sequence of posts, paid placements, bots, account coordination, influencer pickup, and institutional reaction may become a campaign. A campaign tied to deception, targeting, operational timing, or coercive goals may become an operation. Governance becomes confused when it responds to the whole route as if it were only one post.

A sharper definition divides the battlespace into four layers. The artifact layer is the post, image, clip, transcript, prompt output, or ad. The route layer is ranking, recommendation, paid reach, screenshots, influencers, search, messaging, and cross-platform reuse. The authority layer is the moment journalists, officials, platforms, courts, markets, militaries, or schools treat the claim as requiring action. The memory layer is what remains: archives, corrections, labels, logs, appeals, evidence files, and the story later institutions inherit. Most failures happen when the artifact is visible but the route, authority jump, and memory record are missing.

This definition keeps the war metaphor bounded. The battlespace is not a moral label for participants; it is an evidence map for routes. A protester documenting harm, a journalist preserving footage, a foreign service running covert assets, a platform optimizing reach, and a scammer buying engagement can all pass through the same feed. The governance task is to distinguish behavior, authority, and evidence without turning public disagreement into a security category by default.

Narrative as Infrastructure

LikeWar is especially good at showing how narrative stops being soft. A story that spreads can recruit, intimidate, misdirect, normalize, shame, mobilize, demoralize, or make an institution hesitate. Narrative becomes infrastructure when organizations make decisions through it: whether to deploy resources, censor speech, issue a statement, change policy, raise money, launch an investigation, or treat a claim as politically real.

That does not mean that every viral story is false or manipulative. The same network can expose war crimes, document police violence, warn civilians, build mutual aid, and preserve evidence before official systems arrive. The problem is that truth, deception, outrage, grief, entertainment, and status competition share the same routing machinery.

Metrics make this more dangerous. A like count is not a fact, but it feels like social evidence. A trend is not consensus, but it feels like the world speaking. A viral clip is not context, but it can become the first draft of institutional action. A large account is not expertise, but it can trigger the attention of journalists, police, politicians, markets, militaries, and automated systems.

The book's recurring warning is that social media turns public perception into a control surface. If a group can make enough people see a conflict a certain way, it can change the choices available to commanders, parties, platforms, regulators, and ordinary users. The struggle is not only over information. It is over which information becomes operational.

In governance terms, narrative becomes infrastructural at the point where it changes a queue, budget, patrol, market, moderation action, school notice, campaign message, court filing, emergency alert, or military posture. The story does not need to be universally believed. It only needs to move the institution that others rely on.

That is why the site's recurring concern with public memory matters here. The feed is not only a speech channel; it is a record-making system. It decides what appears first, what receives counters, what gets archived, what can be searched, what gets removed, and what remains available to investigators after the political moment has passed. A society that loses the evidence path loses the ability to distinguish public persuasion from public pressure.

The AI Reading

Read in 2026, LikeWar feels like a prehistory of generative conflict. The book already covers bots, sock puppets, trolls, viral manipulation, open-source intelligence, memes, extremist media strategy, and the use of social platforms by states and non-state actors. The AI-era change is that many of these functions can now be cheaper, faster, more personalized, more multilingual, and more visually convincing.

A language model can draft plausible posts for many audiences. An image or video system can supply emotionally charged evidence-like artifacts. A voice clone can make political speech appear to come from someone else. A recommender system can find receptive publics. An agent can run accounts, schedule variants, summarize replies, produce new angles, and route attention to the best-performing story. None of this creates the social media battlespace. It intensifies a battlespace that was already built.

That distinction matters. It is tempting to treat AI as the new origin of information disorder. LikeWar pushes the analysis backward. The deeper condition is a platform environment where engagement is measurable, identity is performative, attention is scarce, and institutional trust is fragile. Synthetic media enters that world as an accelerant, not as a first cause.

The governance question is therefore not only whether a piece of media was generated. A real clip can mislead. A synthetic image can document a real fear. A bot can amplify a true report. A human can lie at scale. The harder question is how platforms, newsrooms, courts, campaigns, militaries, schools, and agencies decide when networked attention has become evidence, pressure, or authority.

That requires separating three AI-era failures. Generated content is the fake image, voice, post, answer, or caption. Generated distribution is the automated testing, targeting, translation, scheduling, and cross-posting that moves the claim. Generated social proof is the synthetic crowd: fake comments, followers, reviews, reactions, or quote-posts that make uptake look organic. Labels on media files help with the first problem, but the second and third require route evidence.

The private-channel version is harder. A chatbot, campaign tool, or agent can test persuasive variants in direct messages, closed groups, customer-service flows, companion interfaces, or workplace channels where public researchers never see a shared artifact. That makes AI contact and bot disclosure, rate limits, campaign logs, consent boundaries, and post-incident retention rules part of information safety, not just privacy paperwork. Public labels are weak when the campaign happens in personalized channels and only the aggregate pressure becomes visible.

Open-Source War

One of the book's most important threads is the rise of open-source intelligence. Networked publics can geolocate videos, identify weapons, trace aircraft, archive deletions, compare shadows, translate local posts, and find inconsistencies before official institutions speak. This is not simply amateur spying. It is a change in who can make claims about reality during conflict.

That is a democratic gain when it exposes hidden violence or counters official denial. It is also a governance problem. Open-source investigation depends on public traces that may endanger civilians, misidentify people, reveal locations, or turn volunteers into participants in conflict. The same skills that preserve evidence can enable targeting, harassment, doxxing, and spectacle.

AI sharpens both sides. Models can help translate, search, cluster, summarize, detect manipulation, and reconstruct timelines. They can also hallucinate links, overstate confidence, launder uncertain claims into fluent reports, or make weak evidence appear polished. The more professional the output looks, the easier it becomes for institutions to skip the slow work of verification.

This is where LikeWar connects to public memory. A conflict is now recorded by phones, platforms, scrapers, archives, dashboards, moderation queues, journalists, NGOs, state offices, and private intelligence firms. The record is abundant but unstable. Posts disappear. Platforms change access rules. Archives are incomplete. Generated media enters the stream. The institution that can preserve provenance, uncertainty, and context will have more power than the institution that merely collects content.

Open-source work therefore needs a modest evidence grammar. Preserve the original file where lawful, record the first known upload, keep hashes or content credentials when available, separate geolocation from attribution, mark confidence levels, and keep corrections attached to the original claim. The goal is not to make every investigator slow. It is to keep speed from turning weak evidence into institutional memory.

It also needs publication discipline. Locating a video is not the same as identifying a person, attributing a sponsor, proving intent, or establishing legality. A responsible OSINT file separates those findings, redacts details that could endanger civilians, marks what was inferred, and records why publication serves the public interest rather than merely feeding the spectacle. The same evidence trail that supports accountability can become targeting material if it is released without harm analysis.

Recursive Reality

The book is also a study of recursive reality: systems changing the world they claim merely to represent. A platform measures what people engage with, ranks more of it, changes what people see, and then treats the changed behavior as evidence of what people want. Conflict actors learn the platform's incentives, produce material for those incentives, watch the public response, and adjust strategy. The public then experiences the adjusted strategy as spontaneous social reality.

The loop is not abstract. A shocking clip provokes outrage. Outrage drives sharing. Sharing draws journalists. Journalists draw politicians. Politicians draw platform enforcement or state response. That response becomes new content. The next side learns what worked. The story becomes not only a description of conflict but one of its instruments.

This loop can make small actors look large, large institutions look weak, falsehoods look socially confirmed, and real harms look like content. It can also make institutions reactive. A ministry, police department, company, school, or court may respond first to the volume of attention rather than the quality of evidence. Once that pattern is learned, attention itself becomes a lever of governance.

AI-era interfaces intensify the loop by compressing observation and response. A model can summarize backlash, classify sentiment, draft replies, generate variants, predict reach, and recommend action. That can help an institution listen. It can also teach the institution to govern through dashboard weather: move when the feed moves, ignore what is not visible, and mistake measured reaction for public reality.

The loop is especially dangerous when interventions become new content. A platform label, government denial, newsroom correction, takedown, legal threat, or military statement can be clipped, framed, and routed back into the same conflict as proof of suppression, panic, or victory. Safety work has to anticipate that every correction is also a signal inside the battlespace.

Governance and Safety

The governance implication is to audit paths, not only artifacts. A useful record for a viral conflict claim should distinguish original source, upload time, edit history, synthetic status, caption context, account identity signals, paid promotion, recommendation exposure, coordinated amplification, platform labels, takedown or downranking actions, appeals, correction reach, and cross-platform reuse. Without that path, the institution is left arguing over screenshots while the system that made the screenshot consequential remains hidden.

For platforms, the safety file should include recommender objectives, ranking changes during crises, ad-delivery records, political and issue-ad archive entries, bot and coordinated-behavior detection, synthetic-media policy, provenance handling, researcher-access decisions, high-velocity virality controls, and post-incident lessons learned. The DSA's risk-assessment, mitigation, recommender, ad-repository, and researcher-access duties are useful because they point at the operating system of attention rather than treating each post as an isolated moderation case.

For newsrooms, courts, public agencies, schools, campaigns, and emergency managers, the operational rule is slower: do not let a platform metric become a fact without provenance. A viral clip may be evidence, but it is not the whole record. A trend may indicate attention, but not consensus. A botnet may amplify a true claim. A generated image may represent a real fear. A real image may be captioned falsely. Governance has to preserve those distinctions when pressure is highest.

For AI systems connected to social platforms, the safety bar should include provenance, account authenticity, rate limits, campaign detection, human approval for political or crisis messaging, logging of prompts and generated variants, and post-deployment monitoring for persuasion at scale. A label on one generated post is weak if the surrounding system can still test thousands of variants, route the best performers into receptive groups, and erase the traces that would let outsiders reconstruct the campaign.

A practical artifact is the conflict-claim docket. For consequential claims, preserve the artifact, source route, known sponsor or unknown-status note, platform and cross-platform path, reach estimates, recommendation or paid-placement evidence, official responses, moderation actions, appeals, corrections, and post-incident change. The docket should include timestamps for first observation, first high-reach amplification, institutional uptake, correction, enforcement, and recirculation. A governance system without clocks will mistake an after-the-fact correction for a remedy that arrived on time.

The minimum safety file should also separate actor, channel, and claim. Who created or sponsored the artifact is one question. How the artifact travelled is another. What the artifact asserts is a third. What an institution did because of it is a fourth. Collapsing those questions is how a real grievance can be dismissed as manipulation, or a manipulative operation can hide inside a real grievance.

Due process also belongs in the safety file. Labels, demotion, account restrictions, ad removals, and takedowns can reduce harm, but opaque enforcement can feed the same distrust that influence operations exploit. Public rules, user notice, appeal paths, researcher access, and audit trails are not niceties; they are part of keeping interventions from becoming fresh evidence for the next loop.

Institutions should drill this before a crisis. Election offices, schools, newsrooms, emergency managers, and civic organizations need named official channels, backup communications, screenshot and link-preservation procedures, escalation contacts at platforms, correction templates, appeal records, and a public after-action note when a false or manipulated claim forced a response. The relevant internal tools are AI audit trails, AI incident reporting, transparency registers, and claim-level source hygiene, not only a faster social media account.

Where the Book Needs Friction

LikeWar is vivid, readable, and case-driven. That is its strength and its risk. The conflict frame can clarify how power uses platforms, but it can also make ordinary politics feel like war. Once every argument becomes a battle and every participant becomes a combatant, democratic disagreement loses some of the vocabulary it needs: persuasion, deliberation, repair, apology, coalition, due process, and shared institutions.

The book also gives less sustained attention to platform political economy than a reader may want now. Engagement design, ad markets, creator monetization, recommender optimization, trust-and-safety labor, data brokerage, cloud infrastructure, app-store governance, and shareholder incentives are not side issues. They shape which conflicts can scale and which harms can be ignored.

Its 2018 publication date matters. It predates the mass public use of large language models, the current synthetic-media policy debate, several major shifts in platform access and moderation, the full public prominence of TikTok in geopolitical anxiety, and the later normalization of AI tools in political communication. That does not make the book obsolete. It means the reader has to update the machinery without losing the core insight.

The most useful update is to read LikeWar less as a fixed map of social media platforms and more as a theory of conflict under measurable attention. Wherever a system makes visibility countable, response rapid, identity performative, and institutional action sensitive to public metrics, the book's logic still applies.

The language of battlespace should therefore be handled carefully. Some actors are coordinated adversaries, some are opportunists, some are ordinary citizens, and some are harmed witnesses trying to be heard. A democracy cannot govern well if it treats every loud public as an enemy operation. The safer frame is evidentiary: map the route, name the behavior, protect lawful dissent, and reserve operational language for operational evidence.

What This Changes

The practical lesson is to stop treating online attention as either mere speech or automatic evidence. It is neither. It is a signal produced by people, platforms, incentives, interfaces, bots, institutions, and sometimes coordinated operations.

For AI governance, that means synthetic-media policy is too narrow if it only asks whether a file is fake. The larger problem is whether a claim can move through systems fast enough to become operational before verification catches up. Labels, provenance, watermarking, takedown processes, media literacy, and bot detection all help, but none of them replaces institutional judgment about when to act, when to slow down, when to preserve evidence, and when to refuse the feed's demand for immediate reaction.

For journalism, courts, schools, campaigns, companies, and public agencies, the key discipline is source separation. Do not let virality collapse witness, evidence, interpretation, and public pressure into one thing. Keep the original artifact, the chain of custody, the context, the edits, the unknowns, the amplification history, and the institutional decision separate enough that each can be challenged.

LikeWar matters because it explains the battlefield before the latest tools arrived. AI can generate the post, fake the voice, summarize the outrage, optimize the target, and automate the reply. But the deeper danger is older: public reality increasingly passes through systems built to reward the material that moves fastest. A society that cannot distinguish attention from evidence will be easy to steer, whether the steering is done by a state, a platform, a movement, a market, or a machine.

The site-level lesson is concrete: every high-stakes attention event needs a source trail and a reaction trail. The source trail asks where the claim came from and how it changed. The reaction trail asks what institutions did because attention made the claim feel urgent. Without both, public memory preserves the spectacle but loses the mechanism.

Operationally, that means building an attention-incident protocol before the next viral crisis. Decide which claims trigger preservation, who can speak for the institution, what evidence must be collected before public action, how corrections will be distributed to the same audience, which records enter the system inventory, and which ad, recommender, or provenance data must be requested from vendors. The point is not to slow every response. It is to stop the feed from deciding what counts as proof.

Source Discipline

This review separates book metadata, author context, defense-journal and academic reception, official legal obligations, regulator findings, standards guidance, and institutional threat reporting. HarperCollins, Google Books, NDU Press, Cambridge Core, New America, and Atlantic Council support the book and author record. The DSA, AI Act, European Commission pages, FTC materials, NIST, C2PA, and EEAS sources support current governance context; they do not prove that any particular platform, model, label, watermark, or enforcement system is effective.

Claims about influence should name the evidence type. A generated artifact, a bot account, a coordinated campaign, a foreign-state operation, a trend, an ad buy, a recommender boost, and a real grassroots reaction are different claims. They need different records: source files, provenance manifests, account metadata, ad-library entries, platform logs, researcher datasets, official attribution, and contemporaneous corrections.

Impact claims need an additional separation. Generation is not reach; reach is not belief change; belief change is not institutional effect. A responsible account should say what is known about creation, distribution, audience, timing, correction, and action, and should mark the unknowns instead of letting novelty stand in for consequence.

Source discipline also means refusing dashboard metaphysics. A trend, impression count, bot score, detector score, virality graph, content credential, or platform enforcement label is a piece of evidence from a particular system, not direct access to public reality. Each should be attached to the method that produced it and the uncertainty it leaves behind.

Foreign-interference and national-security sources need the same restraint. An official threat report can support claims about observed tactics, infrastructure, attribution language, or institutional concern. It does not turn all domestic anger, satire, protest, or error into an operation. This review uses "battlespace" as a route-analysis term, not as a license to militarize civil disagreement.

This review does not claim that any AI system is conscious, divine, or AGI. It treats generative systems, recommenders, agents, and platform metrics as institutional machinery that can shape perception and action while remaining human-built, governed, and contestable.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books