What October 2026 will bring: the AI girlfriend market, forecast

1 · How the last forecast went

Our first forecast, so we tested the method on the past instead

In plain words: We tested the recipe on the last 6 months as if we did not know the answers. It was off by 12.5% on average and landed in range 4 times out of 6.

This is the first edition, so there is no earlier forecast to mark. Instead we did the next best thing: we ran the exact method on the last 6 months as if we did not know how each one turned out, and scored it. On average it missed by 12.5%, 4 of 6 months landed inside its range, and it forecast the direction of the move right 3 times out of 6. That is the honest starting point. From next month on, this section marks our real forecasts.

The backtest, month by month

Each row uses only the data that existed before that month. Worldwide search visits; the US series scored 15.1% average miss.

2 · What the data is doing

The four things we look at before making a forecast

In plain words: Visits are down on a year ago (-16%) and the last three months ran higher than the three before. Name searches are up (+28%). Recent months are getting better, and that is what the forecast follows.

Where traffic is now. September 2026 came in at 8,458,432 visits, +3.7% on the month before and -15.7% on a year ago. The last three months averaged higher than the three before them (+10.0%), which is the momentum we care about most. The 24-month peak was 10,038,844 in September 2025.

Where demand is. People typed the tracked brand names into Google +28.3% more than a year ago, while visits fell, which usually means the big brands are becoming names people search for directly rather than finding through rankings. Google Trends puts interest in “ai girlfriend” at 53 out of 100 for September 2026, +15.2% on the same month last year.

The underlying trend. Fit a straight line through the last twelve months and it slopes -1.3% a month. We do not lean on that line hard, because this market moves in jumps rather than slopes, but it tells us which way the floor is tilting.

3 · What the calendar does

Is October usually a big month or a quiet one?

In plain words: October is usually a quieter month than average (10% below) for interest in AI girlfriends. But the sites’ visits follow rankings more than the calendar, so we show the pattern and do not let it move the numbers.

Five years of Google searches for “ai girlfriend” give every month a personality. The chart scores each month against an average month (100). October scores 90, so it usually runs 10% below average. On the calendar it holds Halloween (31st). The biggest months are June and July; the quietest are November.

The year, month by month

Does the calendar actually move the sites’ traffic? We test that every month rather than assume it. Over the last nine months, a forecast that used the seasonal pattern would have missed by 20.6% on average, and one that ignored it by 11.1%. The pattern did not help: the sites' visits move with rankings and launches far more than with the season. So we show you the calendar and talk about it, but this month's numbers are built on the sites' own recent months. The month the pattern starts earning its place, it goes in, and this paragraph will say so.

4 · The October 2026 forecast

What we expect October 2026 to look like

In plain words: About 8.4M visits in October 2026, about the same as September 2026. Anything between 7.4M and 9.3M would not surprise us.

Worldwide search visits: 8,366,285, in a range of 7,437,628 to 9,294,943. That is about flat against September 2026. US visits: 2,115,935, range 1,864,138 to 2,367,731, which keeps the US at roughly 25.4% of the world.

Why this number: it is a weighted average of the last twelve months, with last month counting for most and each earlier month for less, carried forward, with a range the size of the misses the same recipe made over the last nine months. The weighting is re-chosen every month from that backtest, so the recipe tunes itself to how the market has actually been behaving. No wishful thinking, no model nobody can explain.

Brand and category searches are Google Ads data and run a month behind traffic, so their “last known” is August 2026.

5 · The next six months

Where this goes from here

In plain words: By April 2027 we expect about 8.4M visits a month, so the next six months look flat.

Further out, the range widens, because it should: every month of distance adds a month of things we cannot know yet. By April 2027 we expect about 8,366,285 visits a month, anywhere from 6.9M to 9.9M. Read the middle line as the most likely path and the band as the weather.

6 · Site by site

The ten biggest sites, and what we expect from each in October 2026

In plain words: We expect Live3D and Joyland to be up, and Nastia, OurDream.ai, Romantic AI, AIGirlfriends.ai and BabeChat to be down. Single sites are harder to forecast than the whole market, so their ranges are wider.

Same method, one site at a time, for the ten with the most search visits. We expect Live3D and Joyland to be up on September 2026. We expect Nastia, OurDream.ai, Romantic AI, AIGirlfriends.ai and BabeChat to be down. Single sites are harder to forecast than the market, so the ranges are wider, and a site that just won or lost a big ranking can blow straight through them. Next month this table gets marked.

Worldwide search visits a month per site, DataForSEO; each site’s own last months carried forward with the same method as the market forecast.

7 · How we adjust

What would change our mind, and how the next edition learns

A forecast is only useful if it owns its misses. Here is the loop. On the 1st of every month the new numbers land, this edition’s forecasts get marked against them in public, and the next edition is built on what actually happened, not on what we hoped. When our misses lean one way for a few months in a row, that is the signal to change the recipe, and we will say what we changed and why on this page.

Things that would knock October 2026 off course: a big site winning or losing a page-one position for “ai girlfriend” (the single largest swing factor we have seen), a Google update, a payment-processor or app-store decision that takes a site offline, and a news story that spikes interest for a week and then fades. None of those are in the numbers until they happen, which is exactly why the range exists.

8 · How this is made

The recipe, in full

The data.

Monthly search visits for every tracked site from DataForSEO, 24 months back, summed; brand-name and category search volumes from Google Ads data; five years of interest for “ai girlfriend” from Google Trends via SerpApi.

The forecast.

A weighted average of the last twelve months of each series, with last month counting for 85% and each earlier month for progressively less, carried forward; the weighting is re-chosen every month from a backtest of the series' own last nine months. The seasonal index (five years of Google Trends interest, ratio to a centred twelve-month average) is tested the same way and is not used this month because it made the backtest worse. The range is the method's typical miss over those nine months, widening 10% per month of distance.

The range.

The size of the method’s own typical miss over the recent months, widening with each month of distance.

The calendar.

The seasonal index is shown for context and tested every month; it enters the forecast only if it improves the backtest.

The marking.

Each edition is frozen when published. The next edition scores every earlier forecast for the month that has closed, and keeps a running tally by months ahead.

Where else this appears.

The same forecast feeds the outlook in the AI girlfriend industry report; the month that has just closed is read in full in the AI girlfriend monthly analysis.