“Let’s Bring Back Hope”: An Interactive Discourse Map
Rehumanising Democracy
"Let's Bring Back Hope": An Interactive Map of Where Hope Lands
What happened when a Prime Minister asked the country to hope again
New UK Prime Minister Andy Burnham closed his 20 July inaugural address with a call to Britain to hope again. We analysed 8,086 public comments within the first 48 hours across X, Facebook, Instagram and TikTok, and coded them across 20 DDI behavioural indicators.
Pessimism is not a mood. It is the ground every message now lands on.
What began as a loss of faith in particular institutions has hardened into a loss of the sense of possibility itself. Anger, resentment, and hatred are the visible symptoms, but the deeper condition is subtler and more dangerous — the widespread conviction that the future holds nothing that has not already been tried and failed. A citizen in that condition is not persuadable in the ordinary sense, because persuasion requires believing that evidence not yet seen could still change something.
Prime Minister Andy Burnham deserves genuine credit for identifying this as the critical democratic vulnerability of the moment and for confronting it directly. Much of our politics has responded to public despair by ignoring it or by managing it. To identify the loss of hope as a political problem, and to make its restoration the explicit object of political discourse, is a noteworthy diagnosis.
It is also, for that reason, contested ground. The far right and the demagogues arrived at the same diagnosis some time ago and have built their entire communicative repertoire on it. They speak fluently in the language of cynicism, hopelessness and betrayal, and in speaking it they deepen it. Their advantage is structural.
This research series is an attempt to measure that asymmetry precisely, on a single well-chosen message, at the moment of its reception. Everything that follows is one argument in five movements: what came back, how it was structured, when and where it hardened, and what the pattern asks of leadership.
A hopeful message met a reply that used its own keyword against it
The Democracy Discourse Index scores each comment 0–100 across empathy, civility, epistemic quality and agency. The reception averaged 15.8. Three in four comments fell into the Risk band; exactly one comment in 8,086 scored as Healthy. But the composite alone understates the finding. Split the vocabulary by the stance of the comment carrying it and the message's own word turns out to have changed sides: “hope” appears 800 times inside rejecting comments against 221 inside embracing ones.
Vocabulary by stance
Mentions of each term, split by whether the comment embraced, conditioned on, or rejected the message. Bar length is total volume.
Discourse quality bands
Stance, and its quality
| Stance | Share | Composite |
|---|---|---|
| Reject | 56.4% | 10.8 |
| Conditional | 19.3% | 27.3 |
| Embrace | 14.7% | 20.8 |
| Off-topic | 9.6% | 15.1 |
Note which row scores highest. The conditional cohort — “we'll see”, “show me” — produce better discourse than either the convinced or the hostile. They are the only group still behaving as though evidence could settle something.
Which words travel together
Every term used in at least 70 comments is a node, sized by how many comments contain it and coloured by the language it belongs to. An edge is drawn only where two terms co-occur more often than chance predicts, measured by normalised pointwise mutual information. Drag the threshold to keep only the strongest bonds, search for a term by name, or tap any node to isolate its neighbourhood and read what it pulls with it. Turn on cross-language links to see only the places where one vocabulary reaches into another.
Strongest single associations
The corpus's tightest bonds are procedural and adversarial. “Hope” forms no tight pair of its own — it is spoken through mandate and through grievance, never on its own terms.
Where the languages touch
Each cell sums association strength between two languages; the diagonal is internal cohesion. Click a cell to draw only those crossings.
Six detected communities
Community detection on the same graph recovers six groupings. One is about hope. The other five are mandate, leadership, party continuity, national grievance and policy demands — the ordinary business of politics, which is what the appeal was converted into.
Reading the contact
Care and legitimacy are the two most cohesive languages, and both reach into hope: the appeal was answered in the vocabulary of material demand and of mandate. Dismissal is the most isolated cluster in the corpus — it attaches to almost nothing, because rejection here rarely argues. It asserts.
One dense core, and 37 grievances that connect to nothing
The same analysis at phrase level. A single core of 41 n-grams fuses the slogan itself (bring back · back hope), the mandate demand (general election · call general election) and the leadership vocabulary into one conversation — the message supplied the words for the challenge to it. Outside that core sit nine small clusters and 37 phrases linked to nothing at all.
Clusters by size
Grievances that never become arguments
A comment saying only state pension, or cost living, or mental health raises a real claim but joins no chain of reasoning — nothing else in the corpus picks it up. This is anxiety looking for an object rather than interest making a case. The core conversation is small and mostly procedural; the periphery is large and mute.
A borrowed thread
The core's fourth bundle belongs to a different campaign entirely: polio eradication · restoring support · please restore funding — 41 mentions, tightly interlinked. An organised advocacy push attached itself to a hope message and travelled inside it. Rehumanising discourse has to contend not only with hostility but with capture.
Reception is a property of the hour and the venue, not only of the message
Two findings sit underneath the headline composite. First, the louder the hour, the harsher the reply — so the mean is set largely by the corpus's angriest window. Second, the identical video produced a near-inverted reception depending on where it landed. Neither is a fact about what was said.
Stance across the full window
Rejection thins and embrace roughly doubles as the crowd disperses — 21.0% in the first window, 33.3% in the last.
Volume selects against generosity
Within the peak day, embrace falls from 21.0% to 11.9% while rejection climbs from 47.5% to 61.3%.
Facebook, hour by hour
Embrace share more than halves in seven hours: 23.6% at 16:00, 10.3% by 23:00.
Rejection by platform and period
Darker is higher. Every platform softens into the tail, X by 14 points, but the ranking never changes. Venue effects persist across the whole window.
Which words arrived, and which burned out
Uses per 1,000 comments. Genocide spikes at 64.9 on day one and collapses to 7.9 — an imported campaign that does not persist. Stop and boats move the other way. Mandate, the most constitutional word in the set, falls from 18.5 to 3.8: the legitimacy objection is made loudly and then simply abandoned. “Hope” itself holds steady near 180 throughout — what changes is the company it keeps.
That is what the drift chart shows in miniature. The object of grievance rotates within seventy-two hours while the volume of grievance holds constant — the signature of hyper-scapegoating, a permanent search for whoever can absorb the frustration next.
Communication can diagnose the ailment. It cannot heal it.
The requirement is the rehumanisation of democracy and of democratic discourse: shifting the centre of gravity from anxiety and blame towards empathy, hope and trustworthy civil exchange, backed by real democratic agency rather than mere expression. This corpus is evidence that hope-based communication alone cannot achieve it. The appeal was well made, widely seen, and converted within hours into the vocabulary of mandate and grievance — not because it was poorly judged, but because the ground it landed on completes meaning on its own terms.
What the data does locate is where the remaining possibility sits. The conditional cohort — 19.3% of comments, scoring 27.3 against the rejecters' 10.8 — is the only group still behaving as though evidence could settle something. They are not persuaded and they say so; that is precisely what makes them persuadable. A politics of restoration that addresses the hostile 56% will be met by a repertoire built to absorb it. One that addresses the conditional fifth is speaking to people who have not yet closed the question.
That address cannot be a message. It has to be moral leadership: a recovered culture of moral restraint and, above all, the person who remains good in cynical times. Narrative healing is long work, and it begins with inner work, because a public trained by repeated disappointment can now distinguish authentic care from its performance with considerable accuracy. Leaders who have not done that interior work will be read, correctly, as performing.
Composite scores this low, at this scale, are not a verdict on one speech. They are a reading of the condition into which all speech now arrives — which is why the Democracy Discourse Index measures the reception rather than the message, and will keep doing so.
How this was built
The Democracy Discourse Index measures democratic health through discourse quality, on foundations in critical discourse analysis and deliberative democracy theory. All public comments on the four native posts were captured over 48 hours (9,822), screened to 8,086, and coded against 20 behavioural indicators across four dimensions — empathy, civility, epistemic quality and agency — producing a composite score from 0 to 100. Because the sweep also collects threaded replies to other commenters (1,766 comments, 21.8%), timestamps continue past that window and the observed discourse runs to roughly 72 hours.
Comment text was lowercased and stripped of URLs, handles and punctuation; function words, profanity and the names of private individuals were removed. Terms appearing in at least 70 comments were retained as network nodes (76 terms); co-occurrence was counted at comment level, normalised pointwise mutual information computed per pair, and pairs co-occurring in fewer than 12 comments discarded — leaving 200 pairs above an NPMI of 0.18. The six languages are an editorial classification of the retained vocabulary, not an algorithmic clustering; node positions derive only from association weights. Phrase-level analysis uses two- and three-word n-grams with their own co-occurrence graph and community assignment.
Corpus: 8,086 coded public comments on four native posts, X · Facebook · Instagram · TikTok, 20–23 July 2026. Video reach: 897,000 views on X as at 24 July 2026. Global Centre for Rehumanising Democracy · info@gcrd.org.uk
Volume and quality by window
| Window | Comments | Replies | Composite |
|---|
Read the full policy brief here. This is an independent research initiative of the Global Centre for Rehumanising Democracy, prepared without commission from, or consultation with, any political party or government body.
