Hope in a Low-Trust Democracy: UK Public Discourse in Response to New PM Andy Burnham

How does hopeful political rhetoric perform in a low-trust democratic environment?

Using the Democracy Discourse Index (DDI), our new study analyses 8,086 public responses to Prime Minister Andy Burnham’s “bring back hope” address across Facebook, X, TikTok and Instagram. The interactive analysis examines variation in stance and discourse quality across four DDI dimensions — empathy, civility, epistemic trust and democratic agency — alongside platform effects, temporal dynamics, lexical patterns and evidence of coordinated advocacy. Explore the findings below.

Democracy Discourse Index · Research findings

Public reception of UK Prime Minister’s “let’s bring back hope” address

A multi-platform analysis of stance, discourse quality, temporal dynamics and coordination in public comment on Andy Burnham’s assumption of the premiership.

Corpus

8,086 annotated comments

Period

20–23 July 2026

Platforms

Facebook, X, TikTok, Instagram

Coding

Stance, band and 20 DDI indicators

Volume by platform

Facebook 4,735

X 1,412

TikTok 1,048

Instagram 890

Seven findings · use the controls on each cardSwipe →

Finding 1 · discourse quality

Discourse quality was low across the whole corpus

15.9

/ 100

mean composite DDI score

Band distribution, n = 8,086

Risk 76.1%

Concerning 22.2%

Mixed 1.7%

Healthy 0.01%

Roughly three in four comments fell in the Risk band. One comment in the scored corpus reached the Healthy band. The share in Risk remained above 70% in every three-hour window of the peak day.

Finding 2 · dimension profile

Civility scored highest; empathy scored lowest

Mean scores on the four composite dimensions, each on a 0–20 scale.

Civility & constructive dialogue

5.25

Democratic agency & civic engagement

3.24

Epistemic trust & civic knowledge

3.07

Empathy & perspective-taking

1.04

Indicators near the floor, each scored 0–4

Vulnerability disclosure 0.19 · Narrative use 0.12 · Epistemic humility 0.09 · Source transparency 0.07

The indicators associated with restraint were present; those associated with openness to another position were close to absent.

Finding 3 · stance distribution

Rejection was the majority stance; quality peaked in the conditional segment

Select a stance to read its profile

Reject56.4%4,562 comments Conditional19.3%1,560 comments Embrace14.7%1,190 comments Off-topic9.6%773 comments

Reject

4,562 comments · 56.4%

10.8mean DDI

Finding 4 · platform differentiation

Platform was a strong predictor of tone

X Facebook Instagram TikTok

13.2

mean DDI
composite

83.2%

in Risk band

52.3%

hostility present

Stance composition · 1,412 comments · 17.5% of corpus

Reject 73.7%

Conditional 15.9%

Embrace 7.4%

Off-topic 3.1%

Finding 5 · temporal dynamics

Stance hardened as comment volume rose

Drag the handle through the observation window

1,133

comments in window

17.5

mean DDI

20 Jul
15:00

18:00
peak

21:00

21 Jul
00:00

22–23 Jul
tail

Pre-peak · 20 Jul, 15:00–18:00

Finding 6 · lexical structure

The slogan’s own vocabulary was used most by those rejecting it

“Hope” was the corpus’s most frequent term (1,418 occurrences). Its distribution across stances was uneven.

“hope” in Reject comments

800

“hope” in Conditional comments

381

“hope” in Embrace comments

221

Most frequent phrases

“general election” 366

“bring back” 250

“call general” 119

“british people” 111

“stop boats” 82

Co-occurrence and modularity analysis surfaced six thematic communities: slogan language, election and mandate demands, leadership identity, Labour continuity critique, policy pressures (tax, immigration, public ownership), and a distinct advocacy cluster.

Finding 7 · coordination and corpus profile

A coordinated advocacy layer sat alongside organic reaction

39

Instagram comments from distinct accounts using template variations

21

Equivalent X comments, clustered after the organic peak

A distributed campaign on restoration of UK polio-eradication funding deployed controlled variations of prepared text, mainly on 22–23 July. TikTok showed almost no templating; Facebook showed lower-volume repeated talking points. We assessed this as coordinated authentic advocacy rather than large-scale inauthentic automation.

Corpus profile · proxy indicators

~54%

of usable X location fields were UK-coded

~400

median X follower count; ~7% verified

6–16%

of unique users posted more than once

5:1

male- to female-coded names on Facebook

Demographic indicators are proxy-derived from names, location fields and stylistic markers; they are indicative, not representative.

Interpretation · five-tier trust architecture

Evidence was distributed across five tiers of trust

Select a tier to see the supporting evidence

TIER 5Institutional legitimacymediated by discourse TIER 4Trust in democratic leadersmediated by character TIER 3Social trustmediated by narrative TIER 2Epistemic trustmediated by shared reality TIER 1Meta-foundational trust capacitywillingness to be vulnerable

A message of this kind is received at tiers 5 and 4. The majority of the coded evidence sat in tiers 3 to 1, which messaging alone does not reach.

Implications and limitations

What the findings imply for communication and policy monitoring

01

Single-platform monitoring is biased

Rejection ranged from 40.6% on TikTok to 73.7% on X. Any single-platform reading will systematically over- or under-state opposition.

02

Peak-window sampling overstates hostility

Rejection rose from 47.5% to 61.3% within nine hours, then receded in the tail as quality recovered. Timing of measurement changes the verdict.

03

The conditional segment is the stable audience

At 19.3% of the corpus its share barely moved across the window, and it carried the highest discourse quality of any stance.

04

Advocacy must be distinguished from sentiment

Organised civil-society messaging inserts itself into high-visibility political moments and should not be read as organic opinion.

Limitations: stance and composite scores are coder- and model-derived; demographic indicators are proxy-based; exports capture public comments on specified posts only; coordination detection relies on textual similarity. The window covers four days around the transition.

Full brief and methodology

gcrd.org.uk/insights/bring-back-hope

gcrd.org.uk · info@gcrd.org.uk

01 / 10