Methodology
How the charts on this site are made, and what the numbers and marks on them mean.
AI ranks
6 AI assistants answer the same buying questions weekly. A place averages the last 4 weeks.
People check
Every month we check the top 20 products of each buying question with people. How the check works
You decide
See “Fails the check”, “People disagree” or “Passes the check”? Open the card to see why, with links.
Time to answer one buying question our estimate
Ask an AI 5 min + Read reviews 45 min + Watch videos 90 min + Decide 20 min = 2 h 40 min.
Choose your drink and brewing method to compare coffee setups from our charts. The calculator estimates equipment costs, cost per cup and how long fewer café purchases could take to cover the equipment cost. You can change the assumptions to suit your habits.
Six AI assistants make up the panel. Each Monday, we ask them the same buying questions and record their recommended products and brands.
A chart week runs from Monday to Sunday. We ask the assistants on Monday.
Products earn points when an assistant recommends them, with more points for a higher position. We combine the answers for each question, then combine questions into subcategory, category and overall charts.
The chart averages scores over the last 4 weeks to reduce the effect of a single unusual answer. The monthly check appears beside the ranking and does not change its order.
A new chart uses the weeks available until it has 4 weeks of answers. It is labelled provisional during that time, and badges begin the following week.
The product’s position a week ago and how far it has moved. A dash means it was outside the top 100.
The total weeks a product has spent in this chart’s top 20, including this week. Counting starts with the chart’s first full week.
A chart lists up to 100 places. The top 20 show first; open the rest to see lower positions.
We rank products and their brands from the same answers. Some questions produce useful brand recommendations without consistent product picks; those charts show brands only.
The first week a product is in the top 20 of this chart.
The product at #1 this week was not #1 on this chart last week.
The largest rise into or within the top 20 this week. It must be a climb of at least 10 places, and a chart awards at most one badge.
Back in the top 20 after one or more weeks outside it.
In the top 20 last week and not this week. These products are listed under the chart.
The tiles on the Changes page show leaders in the overall product chart. Larger tiles mean stronger recommendations. Green means a rise this week, red a fall and grey no change.
The Changes page lists the biggest rises and falls across category, subcategory and question charts. Each product or brand appears once, for its largest move. The department lists group these same moves by category.
Chart records track the longest runs at #1, the biggest jumps and falls, and questions whose leaders change most or least often. They use the published positions since each question chart began and show ties where records are equal.
One-week movement records use weeks with fresh panel answers, rather than weeks reconstructed from older data.
Each month, we compare leading products with buyer ratings, review videos, press reviews, owner reports and official recalls. The check focuses on the top 20 and shows its month beside the results. It does not change chart positions.
The marks "Fails the check" and "People disagree" apply to products, rather than their brands. A brand can pass when enough of its products on that page pass; the rule is explained below.
This mark means the check found a clear problem with the product. It appears wherever that product is listed. One rating, video or owner thread alone cannot trigger it; the reasons are:
The AI panel ranks this product much higher than buyers and video reviewers do among the checked products on this chart. We compare them only when enough products have check data. A product that fails the check receives that mark instead.
A product passes when buyers, video reviewers or the press give it strong support, with no conflicting result from the other sources. This needs either agreement across sources or a particularly strong result from many buyer ratings or several agreeing videos. A widespread serious failure reported by owners or an active recall prevents a pass.
The mark summarises the source data in the product’s card. "Fails the check" and "People disagree" take priority. A brand passes when at least half of its products in this page’s top 20 are checked, most checked products pass, and none fails or carries "People disagree".
We have no check data for this product yet. The label does not mean it passed or failed.
An AI model reads review-video transcripts and estimates a score on a five-star scale. This is an AI interpretation of the reviews, rather than a score the reviewers necessarily gave. The card labels it and links to reviews of the exact model.
AI also reads owner discussions to identify serious failures and how widely they are reported. The card shows the count of failure themes and links to the discussions.
This section lists changes near the top of the chart, including rises, falls, debuts, returns and dropouts. Category and subcategory pages also show major moves in their question charts.
On a question chart, each assistant’s column shows where it placed the product in its answer this week. A dot means it did not name the product. On broader charts, the column shows the ranking based on that assistant’s answers alone.
The main chart combines the panel’s scores over 4 weeks, so its leader can differ from an individual assistant’s current choice.
These labels highlight products that one assistant ranks much higher or lower than the other assistants do.
We link to YouTube reviews of products in the top 20, showing each video’s title, channel and views. Short clips link to TikTok videos about the leading products and show their own titles and creators. These lists let you read or watch the sources directly.
Each product recommended in an assistant’s latest answer for a question that week counts as a mention of its brand. For brand-only questions, each recommended brand counts as one mention.
Brand Ocean totals these mentions over the chart’s 4-week window. Within one picture, each dot represents the same number of mentions, so brands mentioned more often appear as larger fish.
Brand Fight compares two brands on questions where both rank in the top 20. Each of six sources can award one point: the AI chart, buyer ratings, AI-read video reviews, press recommendations, owner failure reports and TikTok sentiment. The brand with more points wins; equal scores are a draw.
A fight needs results for both brands from every source except TikTok, which adds a round when both have data. Each source compares eligible models from the shared question charts, counting a model once. Tied results or missing comparable evidence award no point.
The scorecard shows each source’s values and result, using averages before display rounding to decide points. Views and clip counts do not count as approval, and a lack of failure reports does not measure owner satisfaction. The order of rounds changes the presentation, not the outcome.
Prices come from matching offers in US dollars and show the month collected. A single offer links to its seller. Subscription prices show the billing period, and size-dependent prices show the size.
We leave out offers with unclear purchase details or mismatched variants. Open a product’s card to see its price within the subcategory’s range. We do not sell products.
We divide priced products in each subcategory into three equal groups: Budget ($), Mid ($$) and Premium ($$$). Category pages show the median price of each subcategory.
A brand’s tier uses the median price of its products in that subcategory, compared with the same three price groups. It needs at least 2 priced products to receive a tier.
The answers use fixed wording filled in from each chart’s data: the leader, its time at #1, changes over the last four weeks and the monthly check. They update with the chart; we do not generate a new AI-written review for each page.
This section counts the AI answers and source records collected for the chart, including buyer ratings, videos, owner discussions, press reviews and recall checks. It shows the evidence behind the results.
We estimate that researching one buying question yourself takes 160 minutes (2 h 40 min): 5 minutes asking an AI assistant, 45 minutes reading reviews, 90 minutes watching review videos, 20 minutes deciding. We estimate about 2 minutes to read a chart here. These are estimates of effort, rather than measured savings for each visitor.
Our weekly records begin in spring 2026. Recording methods have changed since then, so older results may not be directly comparable with newer ones.
