EggAI-v0.1 · Draft synthetic research · 30 October 2026
The same questionnaire as our forecast-mode report, answered a different way. Instead of a model estimating percentages for the whole audience, 300 simulated UK shoppers each answer every question, using EggAI's own UK model (LBM-UK) and real demographic records from the European Social Survey. Where the two methods agree, the signal is stronger; where they disagree, it is a question for January's data.
Lower-, middle- and higher-income households (by income decile); under 35s and over 65s; households of three or more people; four age groups for drinks. Supermarket segments are not available, because the survey records do not say where people shop.
Panel answers to all 13 survey items, 14 grocery categories and three deal arms: 12,300 model answers in total, each read as answer probabilities rather than a single sampled choice.
Compare with the forecast-mode report, treat agreement as stronger evidence and disagreement as an open question, then check both against January's market data.
Two versions of this study: Forecast mode · Panel mode (this page). Same questions, two ways of answering them.
The panel sees a Christmas bounce; the forecast does not. In every one of 14 categories more simulated shoppers expect to buy more than less over 18–31 December, by +5 points (chilled ready meals) to +15 (cupboard foods, savoury snacks, soft drinks). For the past four weeks the panel shows only a small cut, from −7 (cakes and desserts) to +8 (low/no alcohol). The forecast-mode report found the opposite at Christmas: buying less outweighs buying more in 13 of 14 categories.
Promotions and deals matter less to the panel. Simulated shoppers expect about 21% of their Christmas grocery spend to be on promotion (forecast mode: about 36%). The three matched deals pull similarly: 51% very or fairly likely to move part of the shop for a loyalty price, 52% for a plain 20% cut and 46% for a multibuy. The ranking matches forecast mode, but the gaps are within the model's run-to-run noise.
Big disagreements on premium own label and the inflation warning; agreement on low/no alcohol. Only 7% of the panel say premium own label will be their largest festive-treat spend (forecast mode: 32%); brands lead at 27%. Only 8% had not heard the 7% inflation warning (forecast mode: 47%), and half say it is changing their shopping. On low- and no-alcohol drinks the two methods agree: 24% of the panel expect to buy some, against 25% in forecast mode.
The panel is flat across segments and has weak spots. Most gaps between income or age groups are under 5 points, about the model's run-to-run noise. Opt-out answers are high (9% say they will not buy food for Christmas, 13% don't know), and the discounter answers do not fit together. Read this page alongside the forecast-mode report, not instead of it.
Lower-income households (bottom third), Middle-income households, Higher-income households (top third), Under 35s, Over 65s, Households of 3 or more people
Share cutting volume by category now, and the share expecting to buy more, the same or less of each category at Christmas than last Christmas
Volume change by category through December, and which categories bounce back for Christmas.
Why now: Grocery sales are up about 1% against about 2.5% inflation, so volumes are falling.
For the past four weeks the panel shows a modest squeeze: buying less slightly outweighs buying more in most categories, most for cakes and desserts (−7), fish (−6) and alcohol (−5), while cupboard foods (+3), frozen (+2) and low/no alcohol (+8) edge up. For Christmas it reverses: in all 14 categories more people expect to buy more than less, led by cupboard foods, savoury snacks and soft drinks (+15 each) and fresh fruit and vegetables, meat, dairy and bread (+13).
Forecast mode disagrees on Christmas: it has buying less ahead in 13 of 14 categories (−8 to −17). This is the clearest disagreement between the two versions and the first thing January's volume data will settle.
Higher-income people expect a bigger Christmas lift than lower-income people (meat +16 vs +7), but lower-income people are mainly more likely to answer 'none' or 'don't know', not 'less'.
Lower-income households (bottom third), Middle-income households, Higher-income households (top third), Under 35s, Over 65s, Households of 3 or more people
Expected share of Christmas spend bought on promotion, and which deal type would most make shoppers switch supermarket for part of the Christmas shop
Promotional share of Christmas spend, and switching driven by loyalty prices vs multibuys vs price cuts.
Why now: Supermarket promotion spend is up more than £250m year on year.
The panel expects about 21% of Christmas grocery spend to be on promotion (mean of the stated bands, excluding don't know), with the largest groups at up to 10% (25%) and 10–20% (21%); 15% don't know. In the matched deal test, a plain 20% cut (52% very or fairly likely to move part of the shop) and a loyalty-card price (51%) edge out a multibuy (46%); about 21% are very unlikely to move for any of them.
Forecast mode expects more promotion-led spending (about 36%) and bigger deal effects (73–74% for a loyalty price or plain cut, 60% for a multibuy). The two versions agree on the order: loyalty price ≈ plain cut > multibuy.
Lower-income people are slightly more likely to move for a loyalty price (52%) than over 65s (47%), a gap within noise.
Lower-income households (bottom third), Middle-income households, Higher-income households (top third), Under 35s, Over 65s, Households of 3 or more people
Where shoppers plan to do their main Christmas food shop, the share of Big Four shoppers moving all or part of it to Aldi or Lidl, and the profile of switchers
Discounter Christmas share and the profile of shoppers who switch.
Why now: Lidl is at about 8.6% market share and growing about 7.1%.
13% of the panel expect to spend the most with a discounter this Christmas (Aldi 7%, Lidl 6%), behind Tesco (23%); 9% say they will not buy food for Christmas and 13% don't know. Lower-income people lean most to discounters (17%), higher-income people to Waitrose (10%) and Marks & Spencer (7%).
The panel's discounter answers do not fit together: 21% say all of their Christmas spend will be at Aldi or Lidl, yet only 13% name a discounter as their largest retailer, and even among likely discounter buyers 46% then choose 'neither Aldi nor Lidl'. We show them as answered, but would not use them.
Forecast mode has 19% naming a discounter as their largest Christmas retailer.
Lower-income households (bottom third), Middle-income households, Higher-income households (top third), Under 35s, Over 65s, Households of 3 or more people
For festive treat categories (party food, desserts, chocolates, cheese, mince pies), the split between premium own label, standard own label and brands, by income band
Premium own-label vs branded share in festive categories, by income band.
Why now: Volumes are squeezed, and premium own label grew last Christmas.
In the panel, branded products are the most common largest festive-treat spend (27%), ahead of standard own label (24%); premium own label trails at 7%, and 17% say their household will not buy these treats. Brands rise with income (lower 18%, higher 32%); standard own label falls (30% to 21%).
This is the starkest disagreement with forecast mode, which puts premium own label first at 32%. Last Christmas's published data showed premium own label growing, which favours the forecast-mode reading, but January's data will decide.
Lower-income households (bottom third), Middle-income households, Higher-income households (top third), Under 35s, Over 65s, Households of 3 or more people
Share of shoppers changing behaviour now because of the inflation warnings (stocking up, trading down, no change), and the categories they stock up on
Share of shoppers changing behaviour now, and which categories they stock up on.
Why now: Industry warnings of 7% food inflation in 2027 and the Budget are in the news.
Almost everyone in the panel has heard the warning: only 8% had not. 37% heard it but are not changing anything, while about half say they will act now: 18% stock up and switch, 16% switch to cheaper products or shops only, 9% stock up only and 7% change in other ways. Stocking up centres on tinned, jarred and dried foods (39%), then frozen food (15%).
Forecast mode differs sharply: 47% had not heard the warning and only about 24% are changing behaviour. The category ranking for stocking up agrees (tins first, frozen second).
Higher-income people are the most likely to have heard and ignored it (42%, against 28% of lower-income people).
Aged 18-24, Aged 25-34, Aged 35-54, Aged 55+
Share of Christmas drinks buyers who will buy low/no alcohol, its share of their festive drinks spend, and the split by age group
Low/no share of festive drinks spend by age group.
Why now: Low- and no-alcohol was a visible feature of last Christmas.
24% of the panel expect their Christmas drinks to include some low- or no-alcohol drinks, close to forecast mode's 25%. 21% expect none, 19% don't know yet and 28% do not expect to buy Christmas drinks at all. The age pattern is flat: 21% of 18–24s, 25% of 25–34s, 26% of 35–54s and 23% of over 55s.
Forecast mode has a clear age gradient (33% of 25–34s vs 18% of over 55s); the panel does not. December sales by age, where published, will show which is closer.
Low/no alcohol is one of the few categories where the panel shows more people buying more than less in the past four weeks (+8).
Panel-mode answers were made on 30 October 2026 with no web search: the model answers from its training and each person's demographics. In January we will check them against the same market data as the forecast-mode report, and score the two versions side by side.
| Question | Our forecast now | Checked against | How directly it compares |
|---|---|---|---|
| UK1 · Volume | A Christmas bounce: buying more outweighs buying less in all 14 categories (+5 to +15). Forecast mode predicts the opposite. | Worldpanel (Kantar) volume by category, December | Direction: the key disagreement |
| UK2 · Promotions | About 21% of Christmas spend expected on promotion; loyalty price ≈ plain cut (51–52%) ≥ multibuy (46%). | Promotional share of sales, 12 weeks to late December | Direction: record or not |
| UK3 · Discounters | 13% name Aldi or Lidl as their largest Christmas retailer; the panel's other discounter items are inconsistent. | Discounter value share, 4 weeks to Christmas | Rough level only |
| UK4 · Premium own label | Brands (27%) lead festive-treat spend; premium own label 7%. Forecast mode predicts the opposite. | Premium own-label sales growth | Direction: the key disagreement |
| UK5 · Inflation warning | About half change behaviour now; stocking up (27%) led by tins, jars and dried food, then frozen. | Pantry-category volumes, Nov–Dec | Direction: stock-up or not |
| UK6 · Low/no alcohol | 24% of shoppers buy some low/no for Christmas, flat across age groups. | Low/no alcohol sales, December | Level and age pattern |
This is the panel-mode companion to the forecast-mode report: the same frozen questionnaire, answered by simulated people instead of an aggregate forecast, so the two methods can be compared and both checked in January.
300 people were drawn, weighted by their survey weights, from 1,130 European Social Survey Round 10 respondents in Great Britain (fieldwork 2021). Each person carries only demographics: age, sex, education, region and type of area, main activity, marital status, household size and household income decile.
Segments come from these records: income thirds by decile, under 35s, over 65s, households of three or more, and age groups. Income decile is missing for 35 people, and only 45 fall in the bottom third, so lower-income figures rest on few people.
Every item, category and deal from the frozen forecast-mode questionnaire was asked word for word: 41 questions per person. Lists of retailers, product types and categories were shuffled per person, keeping opt-out answers last; answer scales kept their order.
The question asking which discounter people use was weighted by each person's own chance of shopping at Aldi or Lidl, as its base requires.
EggAI's UK model (LBM-UK, checkpoint E2+org s2, served on our own GPU) answered as each person, choosing one letter. Instead of taking one sampled answer, we read the model's probability for every option and average those across people: 12,300 answers, 99.9% of probability on valid answer letters.
No web search or evidence notes were used. The server was checked against a known-answer test during the run, and 600 answers were re-asked to confirm they were stable (mean difference 3.7 points, normal run-to-run noise).
No calibration was fitted. The panel and forecast versions are compared on every headline in the sections above. Where they agree, the result is less likely to be an artefact of one method; where they disagree, neither is assumed right.
The model was trained to predict how populations answer, not individuals; on person-level tests it is no better than its base model. January's market data is the first real test of both versions.
| Diagnostic | Items | Denominator |
|---|---|---|
| Simulated people | 300 | – |
| Questions answered per person (items × category/deal arms) | 41 | – |
| Model answers (one call per person and question) | 12,300 | – |
| Mean probability on the listed answer letters | 0.999 | 1.00 |
Search the exact question text, base and answer options. Category and deal arms are listed under each question; each respondent would answer once per category, or see one deal arm.
| ID | Method | Base | Question | Options |
|---|
EggAI Research · Draft / 30 October 2026. Exploratory synthetic outputs from the EggAI-v0.1 research harness in panel mode, using the EggAI LBM-UK model; not affiliated with or endorsed by any retailer named. Retailer and brand names are used only to describe answer options. Population records: European Social Survey Round 10 (ESS ERIC), demographics only. Contact [email protected].