COMMENTARY: AfD at 28%, Merz at 21%: A Poll Forces Berlin to Explain Its Priorities

Introduction

From 3 to 5 August 2026, Infratest dimap surveyed 1,297 people for the ARD-DeutschlandTrend and placed the AfD at 28% against 21% for Friedrich Merz’s CDU/CSU. The gap is politically serious, but it is still a poll result rather than an election result. A poll can move a debate without moving a government.

The source also states how the number was produced: telephone and online responses, plus a 2-to-3-point margin of error depending on subgroup size. This commentary takes the signal seriously without turning it into a fictional verdict on Germany’s next government.

The AfD’s 28% is a record in this barometer

The poll gives the AfD a specific national reading

The Alternative für Deutschland reached 28% in the ARD-DeutschlandTrend, a level described in the assigned record as a record for the national voting-intention barometer. It is a dated measurement from one named institute. Twenty-eight percent is a warning, not a mandate.

The figure has political force because it is explicit, not because it is limitless. 28% records declared intentions in this survey; it does not supply a parliamentary majority, a seat projection, or a licence to speak as if an election had already been held.

What The AfD’s 28% is a record in this barometer still leaves open

The AfD’s 28% is a record in this barometer creates a further evidentiary test. The assigned record establishes the poll gives the afd a specific national reading at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. A subsequent official release would clarify whether this is a passing movement or the start of a longer sequence.

For The AfD’s 28% is a record in this barometer, restraint does not weaken the finding; it identifies the next thing that would have to be shown. Later evidence must match the same period before it can revise the present reading. For now, the record allows a firm observation without a forecast.

Merz’s CDU/CSU is measured at 21%

The reading is the lowest since November 2021

The CDU/CSU led by Chancellor Friedrich Merz was measured at 21%, its lowest level since November 2021 according to the assigned material. That is the direct comparison the poll places before Berlin. Merz has a problem. Germany has not cast a ballot.

The result creates pressure on Merz’s political choices, not a settled verdict on his term. Party support in an opinion survey and electoral defeat are different propositions, and the record establishes only the first of them.

What Merz’s CDU/CSU is measured at 21% still leaves open

Merz’s CDU/CSU is measured at 21% creates a further evidentiary test. The assigned record establishes the reading is the lowest since november 2021 at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. A fuller breakdown would show how the reported figure is distributed beneath its headline total.

For Merz’s CDU/CSU is measured at 21%, restraint does not weaken the finding; it identifies the next thing that would have to be shown. That additional detail would add precision, not alter the status of the evidence already cited. The current source is useful because its limit remains visible.

The seven-point gap is a snapshot

The arithmetic is clear but the forecast is not

The difference between 28% for the AfD and 21% for the CDU/CSU is seven points in a poll fielded from 3 to 5 August. The source gives a picture from a short, defined window. Seven points are a gap, not seven certainties.

That photograph changes the debate over priorities, but it cannot determine the next campaign or the next government. The seven-point gap is an observed result; a future outcome would require votes, dates, and evidence that do not appear in this record.

What The seven-point gap is a snapshot still leaves open

The seven-point gap is a snapshot creates a further evidentiary test. The assigned record establishes the arithmetic is clear but the forecast is not at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. The next comparison must preserve the original method and timeframe rather than swap in a convenient new measure.

For The seven-point gap is a snapshot, restraint does not weaken the finding; it identifies the next thing that would have to be shown. Only like-for-like reporting can establish whether the movement has continued. The conclusion remains tied to the defined window in this file.

The fieldwork dates are part of the result

Three days define what respondents were asked

Infratest dimap collected answers from 3 to 5 August 2026. The fieldwork window says when preferences were reported, not what respondents will say after later events or a future campaign. The fieldwork date keeps the claim honest.

A poll is stronger when its time boundary stays attached to it. The three-day fieldwork keeps later developments from being smuggled into the number, while the publication context prevents a temporary measure from becoming a prediction.

What The fieldwork dates are part of the result still leaves open

The fieldwork dates are part of the result creates a further evidentiary test. The assigned record establishes three days define what respondents were asked at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. A documented decision or outcome would be needed before the reported condition could be treated as settled.

For The fieldwork dates are part of the result, restraint does not weaken the finding; it identifies the next thing that would have to be shown. The existing material establishes the issue at hand, not every consequence that might follow it. The appropriate verdict is exact because it stays inside the available proof.

The sample contains 1,297 respondents

Telephone and online answers were combined

The survey used 1,297 people: 773 were interviewed by telephone and 524 responded online. The assigned record therefore identifies both the total sample and the mixed mode of collection. A sample is a lens, not the entire country.

This establishes a method for estimating opinion, not a census of Germany. The sample size makes the result interpretable, while the mixed collection should stop anyone from presenting it as the unfiltered voice of every voter.

What The sample contains 1,297 respondents still leaves open

The sample contains 1,297 respondents creates a further evidentiary test. The assigned record establishes telephone and online answers were combined at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. Future reporting could identify the operational effect that the present item does not describe.

For The sample contains 1,297 respondents, restraint does not weaken the finding; it identifies the next thing that would have to be shown. Until that happens, the source supports a description of the event rather than a claim about its complete impact. The missing detail is a boundary, not a blank cheque for inference.

The margin of error runs from two to three points

Uncertainty changes with subgroup size

The published margin of error ranges from 2 to 3 percentage points depending on the size of the subgroups. That limit is not an afterthought; it is part of what the pollster says the figures can mean. The margin of error belongs in the headline’s shadow.

The margin does not erase a seven-point difference, but it does block a culture of false precision. Statistical uncertainty is the condition of an estimate, and subgroup variation means that not every number carries identical confidence.

What The margin of error runs from two to three points still leaves open

The margin of error runs from two to three points creates a further evidentiary test. The assigned record establishes uncertainty changes with subgroup size at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. A later record would need to identify responsibility, scope, or outcome before this point could be enlarged.

For The margin of error runs from two to three points, restraint does not weaken the finding; it identifies the next thing that would have to be shown. That is the threshold separating a reported development from a completed result. The article keeps that threshold in view.

The record describes a multi-month AfD rise

A broader trend is reported, not fully tabulated

The assigned material says the AfD’s reading fits a continuous rise observed across national German polls over several months. It does not provide every prior wave or a full numerical series. A trend needs more than a slogan about a trend.

That leaves room for a limited conclusion and no more. The reported trend gives context to the August barometer, while the missing complete time series prevents a calculation of an exact slope or a claim about the one cause behind it.

What The record describes a multi-month AfD rise still leaves open

The record describes a multi-month AfD rise creates a further evidentiary test. The assigned record establishes a broader trend is reported, not fully tabulated at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. The material would need a new dated confirmation before the stated risk or promise becomes a measured outcome.

For The record describes a multi-month AfD rise, restraint does not weaken the finding; it identifies the next thing that would have to be shown. No amount of rhetorical force can supply the document that has not been published. Evidence has a timetable of its own.

Migration, the economy, and energy frame the debate

The context names three public issues

The poll arrives amid debate over migration, the economy, and energy policy, according to the assigned record. Those are identified political contexts, not a ranked list of individual reasons for every respondent’s preference. Migration, energy, and the economy do not become one cause.

A sound commentary can name the issues that surround a poll without claiming to read each voter’s mind. Public debate gives the backdrop; individual motivation needs a separate question and evidence that this record does not offer.

What Migration, the economy, and energy frame the debate still leaves open

Migration, the economy, and energy frame the debate creates a further evidentiary test. The assigned record establishes the context names three public issues at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. The next useful report would have to test the mechanism named here rather than merely repeat its language.

For Migration, the economy, and energy frame the debate, restraint does not weaken the finding; it identifies the next thing that would have to be shown. A second description is not automatically a second confirmation. The current record therefore remains the article’s outer edge.

The 2027 budget adds a live test of priorities

The coalition still has to pass the budget

Germany’s coalition government still had to secure its 2027 budget, which provides for a marked rise in defence spending in the assigned record. The fiscal decision lands in the same political season as the poll. A budget fight can sharpen politics without explaining every vote.

That timing makes spending priorities relevant, but it does not demonstrate a direct line from one budget item to one party score. The budget process is a political test; the poll result is a measure of sentiment, not a referendum with one question.

What The 2027 budget adds a live test of priorities still leaves open

The 2027 budget adds a live test of priorities creates a further evidentiary test. The assigned record establishes the coalition still has to pass the budget at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. Further facts may add context, but they must not be imported backward into this dated account.

For The 2027 budget adds a live test of priorities, restraint does not weaken the finding; it identifies the next thing that would have to be shown. The source gives a starting point whose meaning depends on its stated date. Keeping time visible protects the judgment.

Defence spending exposes the cost of choosing

A spending increase competes with other priorities

The record describes a significant planned increase in defence expenditure as Germany debates public spending priorities. It does not supply a respondent-by-respondent explanation of how that proposal affected voting intentions. Defence spending tests priorities before it tests ballots.

This is the substantive pressure behind the numbers: any government must explain what it funds and what it postpones. Defence policy can be central to the dispute without becoming a proven single cause of the AfD’s polling strength.

What Defence spending exposes the cost of choosing still leaves open

Defence spending exposes the cost of choosing creates a further evidentiary test. The assigned record establishes a spending increase competes with other priorities at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. A direct account of implementation would decide whether this stated condition became a practical change.

For Defence spending exposes the cost of choosing, restraint does not weaken the finding; it identifies the next thing that would have to be shown. Until such a record appears, the article can describe the pressure without certifying its final effect. The proof is strongest where it does not pretend to be complete.

The poll identifies its own producer

Infratest dimap is not an anonymous number

The result comes from Infratest dimap for the ARD-DeutschlandTrend, with the field dates, sample, modes, and margin disclosed in the assigned record. That level of attribution distinguishes it from an unsupported viral claim. Method is part of the news.

Naming the producer is not a ceremonial detail. The pollster gives readers a method to assess, while the disclosed design keeps the article tied to what the survey actually measured rather than to headlines built around it.

What The poll identifies its own producer still leaves open

The poll identifies its own producer creates a further evidentiary test. The assigned record establishes infratest dimap is not an anonymous number at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. The necessary next step is a document that answers the question this section leaves unresolved.

For The poll identifies its own producer, restraint does not weaken the finding; it identifies the next thing that would have to be shown. That future evidence may reinforce, qualify, or contradict the present reading, but it cannot be presumed. The existing fact retains its value without being stretched.

The record cannot draw a coalition map

No full party breakdown or seat projection is supplied

The assigned material contains no complete party distribution and no seat projection. It therefore cannot establish a parliamentary majority or a future coalition formula. Coalitions are made after votes, not after charts.

This limit is decisive in a German context where government depends on more than the leading party. The intentions figure can rebalance political pressure, but coalition arithmetic begins only when complete electoral data and actual results exist.

What The record cannot draw a coalition map still leaves open

The record cannot draw a coalition map creates a further evidentiary test. The assigned record establishes no full party breakdown or seat projection is supplied at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. A later assessment would need to establish the wider consequence with the same care as the initial claim.

For The record cannot draw a coalition map, restraint does not weaken the finding; it identifies the next thing that would have to be shown. The available material cannot be upgraded by a more dramatic adjective. Precision is the reason the conclusion holds.

The number changes Merz’s political problem

A low reading requires an answer, not panic

The 21% CDU/CSU reading and its comparison with November 2021 make Friedrich Merz’s task more visible: he must explain the direction of his camp while the AfD leads this barometer. The evidence is a political signal. Merz cannot ignore the number, but cannot read the future from it.

The justified verdict is demanding but limited. Merz cannot dismiss the polling gap as irrelevant; neither can anyone convert it into a final electoral sentence. The next data point, policy choice, and campaign period still matter.

What The number changes Merz’s political problem still leaves open

The number changes Merz’s political problem creates a further evidentiary test. The assigned record establishes a low reading requires an answer, not panic at a stated point in time, but a later dated document would be needed before that point could be extended into a wider conclusion. Only a new source with new facts can move this conclusion beyond its current boundary.

For The number changes Merz’s political problem, restraint does not weaken the finding; it identifies the next thing that would have to be shown. The present record has already supplied what it can: a defined claim, a defined date, and a defined limit. That is enough for a clear judgment today.

Conclusion

The AfD’s 28% and the CDU/CSU’s 21% are not a disguised national election. They are a documented survey result from Infratest dimap, with a known field period, sample, and margin of error.

Their importance is political rather than prophetic. As the 2027 budget and its defence spending debate approach, Berlin must explain its priorities in language that can survive a poll. The chart does not choose a government. It removes the comfort of denial. Berlin’s task is to answer the signal without forging an election.

Signature

Signed Maxime Marquette, columnist

Columnist's Transparency box

Editorial positioning

This commentary is pro-democracy and pro-rule of law in its insistence that public polling be described by its method rather than used to caricature voters or parties. No person named here is assigned a fixed moral category.

The article treats political competition as a demand for accountability, not as permission to inflate a survey into an election. The record’s uncertainty remains part of the argument.

Methodology and sources

The piece relies only on the assigned G2-14 fact record, which identifies Infratest dimap, the ARD-DeutschlandTrend fieldwork, the sample, and the reported margin. It uses the linked reporting for attributed context.

Poll results are stated with their dates, sample, and stated limitations. No seat count, motive breakdown, or future coalition scenario has been inserted.

Nature of the analysis

The analysis argues that the gap changes Berlin’s political incentives, especially around spending priorities. It does not claim that the record demonstrates why every respondent chose a party.

Statements about trends and public debates are kept attributed to the assigned material. The distinction between a measured preference and a future electoral outcome is deliberate.

Sources

Primary sources

Primary reporting is listed here where the assigned record identifies an official document or the originating public statement.

Each entry keeps its publisher, date, and stated scope visible.

• Infratest dimap — ARD-DeutschlandTrend, methodology and results — August 2026
• Infratest dimap — August fieldwork — 2026

Secondary sources

Secondary reporting supplies the attributed context used in this analysis and is not treated as a substitute for a missing official document.

The links retain the source date and the limits of the reporting they summarize.

• osna.fm — ARD-DeutschlandTrend summary — August 2026
• Deutsche Welle — German national polling context — 2026
• Clash Report — reported poll results — August 2026
• Brussels Signal — German political debate — August 2026