We asked one question and Georgia answered. The answer surprised the people who gave it.
One woman looked through the comments under our post and wrote:
That single sentence explains the whole picture. A great many people in Georgia use artificial intelligence — but quietly. Each of them believes it is their own small secret, their own private advantage.
Until we asked, nobody had said it out loud. This survey turned out to be the place where people saw one another.
Health and talking something through come to 59% together. One Georgian in two uses artificial intelligence as a doctor or as a psychologist.
A small survey is often met with doubt. So we compared our result with the most respected polling centre in the world.
Three percentage points sit inside our own margin of error. In other words, our small survey shows what a national American study shows — on a sample ten times larger. That is the best test a method can get.
Source: Pew Research Center, “From diagnoses to treatments: why Americans use AI chatbots for health”, published 25 August 2026.
You will often read that artificial intelligence is used mostly for writing code. That number is real — but it comes from the internal statistics of a particular platform, where programmers are numerous and where coding is steadily moving into a separate working tool. It is a log, not a survey.
When people are asked directly, the picture changes. In Harvard Business Review's 2025 report, which rests on what people actually say, “therapy and companionship” came first — it was second the year before. Personal support is roughly a third of all use and it nearly doubled in a single year. Code sits fifth and eighth on the same list.
So Georgia does not contradict the world. Georgia confirms it, only louder. There, personal support is a third. Here, it is almost two thirds.
Sources: Anthropic Economic Index (a platform's internal statistics); Harvard Business Review, “How People Are Really Using Gen AI in 2025”, Marc Zao-Sanders. The categories do not match exactly — this comparison shows a direction, not an equivalence.
International research is consistent on this: the older the person, the less they use artificial intelligence. In Pew's June study, 44% of Americans aged 18–29 use it for health, against only 17% of those over 65. Three times fewer.
In Georgia the opposite happened. Here is who opened our page:
People over 55 accounted for 81% of all page opens. The over-65s alone took more than half.
And this is not merely because they saw the banner more often. If you count how many of those who saw it went on to open the page, the over-65s still come first: 3.5 opens per 100 views, against only 1.5 in the 35–54 bracket.
And here is what makes that figure twice as interesting. In the campaign settings the age range was set to 18 to 50. We had explicitly asked Facebook to show the ad to a younger audience.
Facebook decided otherwise. Its automatic audience selection treats our setting as a suggestion rather than a rule — and once it saw who was actually responding, it moved the budget itself. As a result, 69% of the impressions went to people who fell entirely outside the range we specified.
So this skew is not ours — it runs against what we intended. The machine spends where it finds a response. It found one in the older generation.
An older person clicks nearly two and a half times more readily than a middle-aged one. Which is exactly what the comments had been telling us:
The reason is simple. A visit to a doctor costs money, and a psychologist costs more. The older generation arrives late, but when it arrives, it stays deeper. Need outweighs mistrust.
One important clarification. We are not claiming that older people in Georgia use artificial intelligence more. ACT Research’s 2025 study (637 respondents) found the opposite: usage among 18–35 year olds is 94%, the highest of any group. We do not dispute that.
Our finding is a narrower one: older people responded far more actively to our offer. And that may be the explanation — younger Georgians already have artificial intelligence; the older generation is only now looking for it.
These figures describe the advertising audience, not the survey respondents themselves — we did not ask for age in the survey. That is exactly why we lean on the "share of viewers who opened the page": it does not depend on how the algorithm distributed the impressions, and it is two and a half times higher among older people.
A digital product in Georgia usually stays locked inside the capital. Not this one: more than half of the campaign went outside Tbilisi.
The question reached Racha and Javakheti alike. Artificial intelligence in Georgia is no longer a capital-city affair — though Tbilisi is still over-represented against the population (about a third of the country lives there, and it took almost half of our reach).
This was the second run of the same promotion. The first time, the same text and the same offer went out on a smaller budget — and almost all of it stayed in Tbilisi, among 25 to 44 year olds.
| Measure | First run | Second run |
|---|---|---|
| Who saw itage and place | 25–44, 98% Tbilisi | 69% aged 55+, 47% Tbilisi |
| Reach | 1,575 | 37,011 |
| Page opens | 21 | 924 |
| Share of viewers who opened | 1.33% | 2.50% |
| Cost per visit | $0.12 | $0.075 |
The older, regional audience responded almost twice as well — and each visit cost 38% less. So it is not simply that the algorithm chose them: the choice turned out to be right.
The two runs also differ in budget and duration, so this is not a clean experiment — but both figures point the same way.
We read all 47 comments and counted which words come back.
One word appeared twice — from two strangers, in two different places. In a comment: “when I need it, it helps me selflessly.” In the survey: “I ask what interests me and it answers selflessly.”
Four of the 47 comments said the same thing. Not that it is bad. Not that it is expensive. But that they could not manage it:
Two walls stand between a Georgian and artificial intelligence. Neither of them is a lack of desire.
The first wall is money. On our own platform even gifted balance is treated as real cash, and people ask for it to be paid out to their card. That is not a complaint — it is the reality of the country.
The second wall is not knowing how. A person is already “friends” with artificial intelligence, yet could not create an account — and apologises for it.
We even offered fifty dollars — and that was not enough either. This is not greed. It means people still do not know how safe this is, or how much of it is theirs. Awareness and trust still come before money.
That is why we give. In eleven days we handed out ninety dollars to people who never paid us a cent — and we never asked them to. Guests are answered for free. Money appears on the balance as a gift.
This is not a marketing cost. It is a digital literacy programme. Since November 2024, when our first user arrived, we have not been buying a market — we have been building one.
An open survey is not a laboratory study and we do not hide that. In exchange, every number here is live and checkable.
Margin of error ±5.3% at 95% confidence. That is the ordinary level of a national poll.
A unique browser key and a daily ceiling per address — two separate layers.
Country is determined by internet address. 336 votes from Georgia, 7 from the diaspora.
Of those who opened the page, 37% also voted. Online surveys usually get 5–15%.
The whole advertising budget was 68.96 dollars. A research company charges 3–15 dollars per answer.
This is not a random sample of the population. It is the voice of the people Georgian-language Facebook reached — 96% of them on a phone.
Do not lean on the small differences — “rights 7.0%” and “code 7.0%” are exactly level and could swap places tomorrow. The large difference (37% against 7%) exceeds the margin of error fivefold — that one is real.
Dunno — the first Georgian multi-purpose AI platform