Skip to content

Jigsaw Prebunking

FreeFirst-line

Publisher: jigsaw.google.com | Type: Non-detector | Cost: Free

TL;DR

Google Jigsaw's prebunking initiative: short (60–90 second) video ads on YouTube, Meta, and TikTok that inoculate viewers against specific manipulation techniques before they encounter them in the wild. The campaign-format inoculation surface that reaches passive audiences at scale; the research-anchored complement to active games like Detektif Deepfake and Bad News.

What it does

Jigsaw Prebunking deploys short video ads (typically 60 to 90 seconds) through YouTube, Meta, and TikTok ad-buying systems. Each ad teaches one manipulation technique (false dichotomies, emotional language, scapegoating, ad hominem, fearmongering) through a brief explanation plus a recognisable example, with the design intent that viewers who later encounter the technique in a real political message recognise it and resist its persuasive effect. The Roozenbeek et al. 2022 Science Advances field experiment with roughly 5.4 million YouTube users documented an approximately 5-percentage-point lift in manipulation-recognition; the Eastern European campaign reached roughly 38 million views.

Jigsaw publishes practical guides, videos, and deployment materials at the Prebunking website; the materials are freely available, the campaign deployment requires a media buy on the underlying ad platforms. The model is the inverse of the active- play inoculation games: where Bad News and Detektif Deepfake expect the user to engage actively for several minutes, Jigsaw Prebunking reaches the user passively in a sub-90-second exposure during the ad slot of a video the user was already watching.

For the fact-checker, Jigsaw Prebunking is the deployment surface that scales: a regional campaign reaching millions of passive YouTube viewers does work no curriculum or game deployment can do at the same audience scale, even though the per-viewer effect is necessarily lighter.

When to use it

  • A coalition campaign in Indonesia is preparing audiences for an election cycle and wants a short-video deployment surface that reaches passive audiences across YouTube, Meta, and TikTok.
  • A media-literacy programme already running curriculum-style with IREX L2D and game-style with Tempo Detektif Deepfake wants a third surface that catches viewers who never engage with active formats.
  • A research-anchored advocacy partner needs the Roozenbeek et al. effectiveness research as the methodological base for funder reporting on prebunking impact at scale.
  • A regional campaign needs deployment infrastructure, ad-creative templates, and effectiveness-evaluation guidance that Jigsaw's toolkit at the Prebunking website provides.

Limitations

Limitations

  • Effects decay rapidly without boosters. The documented half-life is approximately six weeks; a campaign that targets an election cycle should plan booster waves rather than a single launch.
  • Per-viewer lift is modest; the field-experiment literature points to approximately a 5-percentage-point manipulation- recognition improvement, which scales meaningfully across millions of viewers but is light at the individual level. The toolkit does not present Jigsaw Prebunking as a stand-alone solution; it is the campaign-scale layer of a broader programme.
  • The platform layer matters and is not within the campaign operator's control. YouTube, Meta, and TikTok change ad-targeting rules, ad-policy enforcement, and auction dynamics regularly; a campaign budget and reach projection requires reconfirmation against current platform conditions.
  • Localisation matters and is partial. Jigsaw has deployed campaigns in Indonesian (the documented Indonesia case) but the catalogue of pre-built ad creatives skews toward European and US contexts; SEA-specific creative typically requires partner-led development on top of the Jigsaw template framework.

Privacy and threat model

The campaign-creative side of Jigsaw Prebunking is editorial content; deployment is via the standard YouTube / Meta / TikTok ad systems, which carry those platforms' standard targeting, data-collection, and reporting practices. The campaign operator is the ad-account holder and inherits the platform's data- handling terms; the audience experiences the ads as part of their normal feed.

The threat model that matters here is platform-level rather than data-flow: campaign creatives may be paused, removed, or restricted by the platform's ad-policy enforcement in ways that are inconsistent across regions and that can shift mid-campaign. Operators in surveillance-environment countries should consider that ad-targeting metadata is visible to the platform; for Sri Lanka and Laos, that visibility is one of the considerations in deploying any ad-funded campaign.

Country and platform applicability

  • Indonesia: documented Jigsaw deployment (REGIONAL- CASES, Indonesia by reference; the Indonesia campaign cited ). Pairs with Tempo Detektif Deepfake and IREX L2D for a complete three-surface Bahasa programme.
  • Laos: Lao-language ad creatives are not documented ; deployment would require partner-led creative development, and the platform-reach question in Laos is itself uneven (lower YouTube reach than Indonesia or the Philippines).
  • Malaysia: no documented Malaysian-specific deployment cited ; the Indonesian Bahasa creative could plausibly extend with Malay adaptation.
  • Philippines: no documented Filipino-specific deployment cited; partner-led adaptation is the operational route and could be paired with the #FactsFirstPH coalition.
  • Sri Lanka: no documented Sinhala or Tamil deployment cited; partner-led adaptation through Hashtag Generation, Watchdog, or DRI could plausibly lead.
  • Thailand: no documented Thai-specific deployment cited; Cofact Thailand or AFP Fact Check Thailand could plausibly lead.

Platform applicability: YouTube, Facebook, Facebook Groups, TikTok as the documented deployment surfaces. The campaign reach in any country is a function of the platform mix in that country — high YouTube reach in Indonesia and the Philippines, lower in Laos.

How to access

Visit prebunking.withgoogle.com (Jigsaw's Prebunking website) for free practical guides, the existing video catalogue, research summaries, and deployment templates. Campaign deployment uses the standard ad-buying interfaces of YouTube, Meta, and TikTok and is gated by the deploying organisation's ad-account access on those platforms.

Cost (current as of 2026-05)

The Prebunking-website resources, guides, and existing video templates are free. Campaign deployment cost is the media buy on the underlying ad platforms, which scales with audience reach and platform CPM rates; a meaningful regional campaign budget typically runs in the tens of thousands of dollars to reach the millions-of-views level documented in the published case studies.

Quickstart (deployment-side)

  1. Review the Prebunking website's research base, video templates, and deployment guides; identify the manipulation techniques most relevant to your audience moment.
  2. Decide whether to deploy an existing creative (faster, less localised) or commission partner-led creative work in the target SEA language (slower, more effective in context).
  3. Confirm the ad accounts and budgets needed on YouTube, Meta, and TikTok; account access and policy approval can take time on each platform.
  4. Plan the campaign waves; a single launch is less effective than booster waves spaced to maintain the recognition lift over the campaign window.
  5. Pair the ad-deployment surface with the active-engagement surfaces from the rest of the cell (Tempo Detektif Deepfake and IREX L2D) so passive viewers who get curious have a path into deeper engagement.
  6. Plan effectiveness evaluation per the published Roozenbeek et al. methodology if institutional reporting requires it.

In the toolkit's workflow

Source-history and behaviour pillar non-detector audience- engagement deployment per Architectural Anchor 1. Sits in 2A.3 as the alternative entry covering the campaign-scale inoculation surface: the deployment shape that reaches millions of passive viewers at the cost of lighter per-viewer effect, complementing the active-engagement game and curriculum surfaces elsewhere in the cell.

Standard combinations (deployment combinations):

  • With Tempo Detektif Deepfake at 2A.3 – Jigsaw reaches passive audiences with short video; Tempo's game catches the audience that wants active engagement. Together the two surfaces cover the engagement spectrum in Bahasa.
  • With IREX L2D / Gali Fakta at 2A.3 – Jigsaw campaigns drive awareness and curiosity; IREX's curriculum and game absorb the audience that follows the curiosity into deeper learning.
  • With Bad News / Harmony Square at 2A.3 – as the cross-region research benchmark; Jigsaw and Bad News are the two largest published prebunking research bases.
  • With Meedan Check and Kalimasada at 2B.1 when campaign-level audience engagement generates downstream tipline volume that needs editorial processing.

Decision-tree references: this is not a verification routine; the decision trees in T1–T7 do not invoke Jigsaw at runtime. The campaign sits in the editorial-strategic layer that shapes audience preparation upstream of any specific claim handling.

This card carries no detector signal class: Jigsaw Prebunking deploys inoculation campaigns; it does not produce a verdict on any specific piece of suspect content.

Override notes

Override notes

  • Indonesia campaign deployed (b4-score-2-Indonesia- campaign): the change-log record cites the Indonesian campaign as the SEA-region anchor. The lighter B4 score (2 of 5) reflects that this is one Indonesian campaign rather than a sustained Indonesian deployment; partner-led extension and reinforcement is the operational route to deeper SEA reach.

Sources

  • Jigsaw / Google. Prebunking resources, video templates, and deployment guides. Jigsaw, 2023–2024. prebunking.withgoogle.com.
  • Roozenbeek, J. et al. Psychological inoculation improves resilience against misinformation on social media. Science Advances, 8(25), 2022. doi.org/10.1126/sciadv.abo6254.