Ahead of President Trump’s 2026 State of the Union, Statt published a full predicted AI-generated transcript of what we expected to hear from the podium.
We built it using Statt’s automated policy speechwriting tools and predictive analytics that synthesize upstream signals across our policy datastream.
After the speech, we graded our prediction against the official transcript. Not on cherry-picked quotes, but on the questions policy teams actually care about: Did we anticipate the agenda? Did we capture the framing? Did we weigh the right issues? Did we correctly predict what the speech was trying to accomplish?
Ultimately, we were strongest on policy architecture and narrative framing. Where we fell short was the performative layer of a prime-time speech, especially guest stories and ceremonial moments.
How we generated the predicted transcript
Statt’s AI-generated policy speechwriting tool generated a first-draft State of the Union transcript in Trump’s voice by grounding the model in upstream signals and reference material, including prior Trump remarks, past State of the Union addresses, public statements, and policy analysis from aligned research ecosystems.
How we evaluated accuracy
We conducted a thematic comparison between our predicted transcript and the official transcript of President Trump’s 2026 State of the Union. Rather than scoring word-for-word accuracy, we used an “LLM as Judge” approach with Claude (Anthropic) to identify whether we correctly anticipated:
- The major policy pillars
- The sequencing and narrative architecture
- The framing and rhetorical strategy
- The relative prominence of each issue
- The likely policy follow-through embedded in the speech
Claude reviewed both transcripts side by side and graded approximately 20 thematic categories spanning economic policy, immigration, foreign policy, domestic policy, and rhetorical and performative elements, such as guest stories and ceremonial moments.
Prediction results: How close did we get?
Below is the simplified scorecard we used to grade performance. It’s designed to answer one question: Did we predict the agenda and framing correctly, and where did we diverge?
What we got right
The strongest takeaway is that we consistently captured the policy architecture of the speech, the big buckets Trump chose to prioritize, and the way he framed them, even when we missed some of the stagecraft.
Border and immigration
We nailed this as a major centerpiece, and the actual speech hit almost all of the same notes we forecast: record-low crossings, zero interior releases, heavy emphasis on ICE enforcement, sanctuary city attacks, the SAVE Act and voter ID, an end to catch-and-release, and the DHS funding fight with Democrats. We also correctly predicted the “invasion” framing and the attack on the Biden-era app. We missed some specific proposals and named stories, but the broad structure and posture were highly accurate.
Trade, tariffs, and the Supreme Court ruling
This was one of the most specific and most consequential predictions we got right. We anticipated that tariffs would be a major topic, that the Supreme Court would have recently ruled against IEEPA tariff authority, and that Trump would frame it as an obstacle he would work around using alternative legal authorities. The actual address confirmed the same storyline: existing deals hold, alternative statutes exist, and the agenda continues. We were also close on the defiant posture. The main difference was tonal texture — the real speech was less pointed at the Court than our predicted version.
One Big Beautiful Bill and tax policy
We correctly forecast the package Trump treated as proof of “wins”: no tax on tips, no tax on overtime, no tax on Social Security, an expanded child tax credit, school choice provisions, the $1,776 Warrior Dividend, and Trump Accounts. The speech covered these in a similar “money back now” framing. We missed a handful of specific details and story attachments, but we were directionally right on both content and positioning.
Energy dominance and permitting
We correctly predicted energy as foundational to the economic story, not just a standalone policy section. The address used energy as the master explanation for inflation relief, manufacturing momentum, and national security. We also predicted that permitting reform would show up as a major “unlock” narrative, and the actual speech treated it the same way: as the enabling constraint behind energy, housing, infrastructure, and technology.
Deregulation
We correctly forecast deregulation as a headline accomplishment and as part of the broader “turnaround” claim. Where we diverged was that our predicted transcript included specific ratios and savings figures that did not appear in the actual speech. The lesson is that the speech leaned more on the theme than on the math, which is useful guidance for calibrating specificity in future predictions.
Iran and Operation Midnight Hammer
We correctly predicted this would be a signature national security segment, including the operation name, June timing, the decades of failed diplomacy framing, and the posture of negotiations from strength. We also predicted Witkoff and Kushner as the negotiators, and the actual speech confirmed it. The real address added more narrative detail and historical callbacks, but the core sequencing and intent remained the same.
Venezuela and Maduro
We correctly predicted that a military operation against Maduro would be presented as a major triumph. Where we were less accurate was in scale and emphasis. The actual speech treated Venezuela as a defining emotional and narrative centerpiece, with far more detail than we included. The prediction was directionally right, but we underweighted how central this would be to the night.
Healthcare and drug pricing
We were also directionally right on healthcare and drug pricing. Our draft anticipated a direct-negotiation posture and a push to frame savings as flowing to patients rather than to intermediaries. The actual speech leaned into the same conceptual story, even if the packaging and specific examples differed from what we wrote.
AI and technology
We correctly forecast that AI would be treated as a national power issue, tied to competition with China and the infrastructure required to scale. While we missed some of the most specific policy hooks in the actual address, the underlying frame matched: AI as a strategic priority, with permitting and buildout capacity as enabling constraints.
Military strength and recruitment
We were right in the direction that the speech would use military strength and recruitment as proof points in the broader “turnaround” storyline. The actual address leaned more explicitly into recruitment success and readiness language than our draft did, but we anticipated that this would be presented as evidence of restored national confidence.
Closing rhetoric and overall arc
Finally, we captured the rhetorical bookends of the address, including the “Golden Age of America” framing and the patriotic, worker-focused cadence of the close. We were stronger on the broad arc and tone than on the specific televised vignettes that carried the ending in the actual speech.
What we missed
The misses are instructive because they clarify what is hardest to forecast from policy documents alone.
The “television spine” of the speech
The biggest miss was the speech’s heavy reliance on guest vignettes, named individuals, and ceremonial moments that carried the night from segment to segment. Our predicted transcript was policy-forward. The actual speech was policy plus a series of emotional set pieces designed for the room and the cameras.
Culture war and social issues
We underpredicted the extent to which cultural issues would be a meaningful section of the address. These moments often respond to the media cycle and the principal’s instincts, which makes them less legible in traditional policy text signals, but they still consume real speech time and shape the overall tone.
Fraud and corruption
We missed the fraud and corruption angle that became a notable section in the actual address.
Foreign policy enumeration and claims
We were directionally right on the major foreign policy posture, but missed how the enumerated and scorecard-like parts of the foreign policy section would be.
Specific policy announcements and surprise elements
We missed several surprise crowd moments that are difficult to infer from policy data alone but are frequently present in televised addresses.
How Statt helps teams predict policy narratives before they happen
Statt can not only anticipate the structure and framing of a national address like the State of the Union, but we can also apply the same AI-capabilities to the moments that matter to policy teams every week, including:
- Generating first drafts in the voice of a policymaker or executive
- Running “murder board” prep for hearings, testimony, and high-stakes meetings with mock Q&A
- Producing probability-weighted scenario memos that outline likely policy moves and the risk those moves create for your business or clients.
See predictive intelligence for your priorities
If you want to see how Statt generates predictive readouts and fast deliverables for your issues, request a custom demo. We’ll tailor it to your policy portfolio, jurisdictions, and the cadence your team runs on. Request a custom Statt demo.