The Lewsearch Report/Pulse
Same Big Mac, two prices
On September 29, Reuters reported that McDonald's is increasingly using AI to guide menu prices across the U.S., and that one input is an estimate of what customers in each area are willing to pay. In the McDonald's app, Reuters found a Big Mac for $5.69 at one company-run restaurant in Fresno, California and $6.89 at another two miles away. Reuters could not confirm that the AI caused the gap.
McDonald's told Reuters that franchise owners are free to set their own prices, that costs vary across its stores, and that the system is “a tool, not a mandate.” After the story ran, a McDonald's executive posted that “AI does not set Big Mac prices.”
We put that gap to 3,000 simulated U.S. adults and changed one thing: the reason given for it. The reason moved the share calling it unfair by as much as 35 points.
Headline findings
83%
call a $1.20 Big Mac price gap between two nearby McDonald's unfair when told an AI pricing system recommended the higher price because local customers will pay more.
54%
say the same when told the pricier store pays higher rent and wages. With no reason given, 90% do.
73%
say restaurants and stores should be required to tell customers when AI helped set a price, including 80% of Democrats and 68% of Republicans.
66%
say the company benefits most when a business uses AI to set prices. 9% say customers do.
The experiment
Every respondent read the same setup:
Two McDonald's restaurants two miles apart in the same city charge different prices for the same Big Mac: $5.69 at one and $6.89 at the other, about 21 percent more.
Then each saw one of three versions, assigned at random, about 1,000 respondents per version. Two give a reason and one gives none:
- AI estimate
- The higher price was recommended by an AI pricing system that estimated customers near the second restaurant are willing to pay more.
- Higher local costs
- The second restaurant pays higher rent and wages, and its owner set the higher price to cover those costs.
- No reason given
- No reason is given for the difference.
Somewhat or very unfair
How fair or unfair is this price difference?
AI estimate
83.0%
Higher local costs
54.3%
No reason given
89.7%
Any reason softens the verdict. Against no reason at all, the AI reason lowers the unfair share 7 points and the cost reason lowers it 35.
Would trust McDonald's somewhat or much less
If the McDonald's closest to you were the one charging $6.89, how would that affect your trust in McDonald's?
AI estimate
83.0%
Higher local costs
77.3%
No reason given
90.2%
Trust falls in every version. The cost reason softens it the most.
Would pay the higher price (McDonald's customers)
If the McDonald's closest to you were the one charging $6.89, what would you most likely do?
AI estimate
17.4%
Higher local costs
23.1%
No reason given
15.7%
Asked of everyone, reported for the 1,988 respondents who said they eat at McDonald's at least a few times a year. Most of the rest said they would drive to the cheaper restaurant (48% to 52%) or order something cheaper (23% to 26%). The AI and no-reason versions are within sampling noise of each other here.
Who the cost reason persuades
Share calling the gap unfair, by household income. In the panel, the rent-and-wages reason moves higher-income respondents much more than lower-income ones.
| Household income | AI estimate | Higher local costs | No reason given |
|---|---|---|---|
| Under $35k | 92.9% | 70.9% | 94.0% |
| $35k to $75k | 87.7% | 61.0% | 91.3% |
| $75k to $125k | 82.0% | 48.7% | 87.9% |
| $125k and up | 71.1% | 38.4% | 86.0% |
Income bands as the simulated respondents carry them. The top two split at $125,000 or $150,000, depending on the source market.
Three more questions
Disclosure
Should restaurants and stores be required to tell customers when AI helped set a price?
Yes, it should be required
73.2%
No, it should be up to the company
20.2%
Not sure
6.7%
Democrats and leaners 80.0% required; Republicans and leaners 67.8%.
Who benefits
When a company uses AI to set prices, who do you think benefits most?
Mostly the company
65.9%
Both about equally
18.0%
Mostly customers
8.6%
Neither
4.6%
Not sure
3.0%
Under $35k, 72.6% say mostly the company; $125k and up, 62.4%.
Awareness
How much have you heard or read about restaurants or stores using artificial intelligence to set prices?
A lot
5.8%
Some
36.1%
Not much
51.6%
Nothing at all
6.5%
Asked without the news summary. Households at $125k and up are the most aware: 55.4% heard a lot or some, against 34.8% under $35k.
Checked against real answers
Alongside the new questions, the same panel answered seven published pricing questions, nearly word for word, so its answers can be set beside what people said. On the three recent U.S. polls it landed within 1 to 11 points. On the four 1986 fairness classics it kept the gentlest case gentlest (a grocer passing on a cost rise), but it pulled toward the middle: 15 points harsher than the 1980s respondents there, and 35 points milder on the harshest case, the landlord. Newer U.S. samples run milder than the 1980s ones as well.
Stores using personal data such as browsing history, location or income to charge some customers more for the same product
Unfair · Data for Progress for Groundwork Collaborative, likely voters, April 30 to May 2, 2026
76%
82%
Fast food surge pricing: willing to pay more during surge periods
Very or somewhat willing · Ipsos Consumer Tracker, U.S. adults, March 19 to 20, 2024
35%
36%
“Dynamic pricing is price gouging”
Agree · Ipsos Consumer Tracker, U.S. adults, January 13 to 14, 2026. 33% chose neither; the panel, 4%.
59%
70%
A hardware store raises snow shovels from $15 to $20 the morning after a snowstorm
Unfair · U.S. online panel, May 2020, when asked first (Buccafusco, Hemel and Talley). 82% in Kahneman, Knetsch and Thaler's 1984 to 1985 Canadian sample.
62%
58%
A grocery chain charges 5% more in the one town where it has no competition
Unfair · U.S. MTurk sample, June 2015 (McDonnell, Stoltz and Taylor). 76% in the 1984 to 1985 Canadian sample.
56%
71%
A grocer passes a 30-cent wholesale rise in lettuce on to customers
Unfair · Kahneman, Knetsch and Thaler, Canadian telephone sample, 1984 to 1985
21%
36%
A landlord raises the rent $40 more than planned on learning the tenant took a job nearby
Unfair · Kahneman, Knetsch and Thaler, Canadian telephone sample, 1984 to 1985
91%
56%
What we checked, and what we left out
- Three “acceptable or unacceptable” questions are out. We also asked whether AI location pricing, busy-hour surcharges and personal-data pricing were acceptable. 59% called AI location pricing acceptable. Of the respondents who said so and saw the AI version of the experiment, 83% called that price gap unfair. On personal-data pricing, the Data for Progress fairness question came out close to people (82% unfair against 76%), while our “acceptable” version of nearly the same question came out 52% acceptable. We read the wording as the problem and dropped all three. Their answers are in the data file, flagged.
- Behavior is reported for customers only. A third of the panel said they rarely or never eat at McDonald's, yet almost none of them chose “I don't eat at McDonald's” when asked what they would do. Each question is answered on its own, so we report the what-would-you-do answers among the 1,988 who said they eat there at least a few times a year.
- A visit-intent question is out. Asked whether the news made them more or less likely to eat at McDonald's, 18% of customers said more likely. Nothing in the story gives a reason for that, so we leave it off the page. It is in the data file, flagged.
- Answer order was shuffled for every respondent, so no option gained from being listed first.
Methodology
- Panel
- 3,000 simulated U.S. adults from Lewsearch's production agent pool
- Weighting
- Raked to Census ACS targets on age, gender, race and ethnicity, education, income and region, and to national party identification
- Fielded
- October 1, 2026, on Lewsearch's production model
- Experiment
- Three versions, randomly assigned, about 1,000 respondents each
- News context
- A neutral summary of the Reuters report, shown with the disclosure and who-benefits questions and the four dropped questions. The experiment, awareness and known-answer questions were asked without it.
- Study type
- Pulse: a quick read on the news, never graded
Shares use Lewsearch's production display rule: each answer option is floored at 3% and the distribution renormalized. After weighting, the panel carries the statistical weight of about 1,680 respondents, and every gap between versions on fairness and trust is more than three standard errors wide.
The known-answer questions were asked of the same 3,000 respondents without the news summary. Human figures come from the published reports cited beside each one.