How We Pick
This page is what makes the word “best” defensible anywhere on this site. It is also where we state plainly what we do not do.
By Mike M. · Published November 6, 2026 · Last updated November 6, 2026
We do not physically test pool cleaners
We do not own a test pool. We do not own these machines. Nothing on this site is written from first-hand use, and no page here claims otherwise.
You will not find the phrases “we tested”, “field-tested”, “in our test pool”, “after six weeks” or “our measurements showed” anywhere on this site, because none of them would be true.
We are saying this on the most prominent trust page rather than burying it, because it is the thing a reader most deserves to know and because it shapes everything else. It is also, frankly, the editorial line we would rather compete on.
The leading independent site in this category claims to have tested thirty-plus robots — and on the category pages we examined, published no test data, no measurements, no methodology and no cons at all. An honest synthesist who shows their sources and names who each product is wrong for is more useful than an unverifiable tester who publishes neither.
The four evidence types we do use
Every factual assertion on this site traces to one of four sources, and every spec row on every product page is tagged with which one.
| Tagged as | What it covers |
|---|---|
| manufacturer | Rated pool size, suction GPH, filter micron, battery capacity, runtime, charge time, cable length, warranty, brush material. Taken from the brand’s own published specifications, with the brand named in the Sources section of the page. |
| retailer listing | Figures published on the Amazon listing for that exact ASIN — coverage claims, runtime, included accessories, model codes, and whether a unit is new or factory-renewed. Tied to a specific listing, not carried across between variants. |
| owner reports | Real-world durability patterns, recurring faults, what breaks and when. Always described as owner reports, never as our own experience, and never quoted as a testimonial. |
| not published | Neither the manufacturer nor the retailer publishes the figure. We write that rather than estimating one. This tag appears more often than you would expect — including on machines costing over $1,000 — and for a buyer comparing on paper, an unpublished figure should be treated as an unknown rather than as “probably fine”. |
Where evidence does not exist, the page says so. That sentence is better content than a fabricated number, and it is the one competitors never write.
The scoring rubric, with its weights
Every product carries five metric scores out of 10 and a weighted overall. All five are derived from the published figures above — they are arithmetic on specifications, not test results.
Here is the whole method, so you can disagree with us by checking it.
| Metric | Weight | Derived from |
|---|---|---|
| Debris capacity | 20% | Published basket or bin volume, plus intake width where published. 5 L and above scores 9+; 4 L around 7.5; unpublished or visibly small capacity scales down. This is the single best predictor of leaf performance. |
| Surface coverage | 25% | How many of floor / wall / waterline / water surface the machine publishes. Floor only scores around 4; floor and walls 6.5; adding a published waterline pass 8.5; adding surface skimming 10. A wall climb is not counted as a waterline pass. |
| Filtration | 20% | Finest published micron rating. 3 micron scores 10; a 10-micron layer 8.5; 150-180 micron 6; unpublished scales to 4-7 depending on whether a fine option exists at all. Replaceable ultra-fine cartridges count in a machine’s favor. |
| Runtime & coverage | 15% | Published runtime against published rated pool area — whether the machine finishes its rated pool in one cycle. A corded machine with no runtime limit scores 9.5-10; a 90-minute 860 sq ft machine around 5. |
| Value for spec | 20% | The published spec set against the live price band, including consumable costs and whether a replacement battery is sold. This is the metric that moves when a price moves. |
The score is visible content only. It is never emitted as structured data. Marking an editorial score up as a review rating in JSON-LD would contradict the first section of this page, and it is a structured-data policy risk. There is no star-rating or aggregate-score markup of any kind in this site’s structured data — only Product and Offer nodes, and the Offer only when a live price exists.
Two deliberate consequences of the weights. First, surface coverage carries the most weight, because a machine that does not reach the surface you care about cannot be fixed by any other spec. Second, no metric is scored on brand reputation, review counts or star ratings — we hold none of those and will not manufacture them.
How prices work, and why they vanish
Every price on this site comes from the Amazon Creators API. There is no price field anywhere in our content files — it is structurally impossible for a human to type a price into a page here, which is deliberate.
Each price carries the date it was checked. Right now that is November 6, 2026, covering 67 of the 37 products and parts in our registry.
Prices expire on their own. If the price data is more than 48 hours old, every number on the site disappears and the buy buttons revert to “Check price on Amazon”. We do not fall back to a remembered figure, a cached figure or an approximate band. The failure mode is silence.
That matters because Amazon prices move constantly and a stale price is a small lie that costs a reader real money. It also means that if you see a number here, it is current or it is gone.
One exception worth naming: where we discuss manufacturer list prices or price bands in prose — “around $500”, “roughly $26 for four” — those are editorial context, dated to this page’s last update, and the live figure in the buy button is always the authority.
How products get onto the site at all
The sequence, in order:
- Query research. We start from what people actually search for in this category — the problems, not the product names.
- Candidate selection. Real, currently-sold models that address those problems, across every brand rather than one.
- Live verification. Each model is resolved against the Amazon Creators API and checked for a live listing with a buyable offer and a real product image. A model we cannot verify as currently sold does not get a page.
- Substitution where needed. Where a model we planned to cover is not currently sold, we substitute the nearest in-catalog machine from the same brand and tier and change the page to match the machine — rather than publishing a page about something you cannot buy. One page says so explicitly.
- Specs recorded with sources. Every figure tagged per section 2.
- Scored by the rubric. Per section 3.
- “Who should skip this” written for every product. Mandatory, including on the machines we recommend most.
No brand pays for inclusion, for a ranking, or for a review. No manufacturer sees a page before it publishes. We have no relationship with any brand on this site beyond buying from the same retailer you do.
What we will not publish
- Testimonials or customer reviews. We have none, so there are none on this site. Not a grid of plausible-looking quotes, not star ratings aggregated from somewhere else.
- Manufacturer error-code tables. They are model-specific, change between hardware revisions, and the versions circulating online are frequently copied from the wrong model. We teach the four fault categories instead and link each brand’s own support page.
- Specific warranty terms. They vary by region, listing and whether stock is new or renewed, and a stale warranty table is the kind of error a reader makes a $900 decision on. We publish the list of questions to ask instead.
- Invented figures of any kind. No estimated runtimes, no approximate decibel readings, no “about 90% coverage”, no made-up micron ratings.
- Author credentials we do not hold. No certifications, no “expert panel”, no claimed trade background. The about page says exactly who writes this.
- Seasonal deal pages with fabricated discounts, countdowns, or fake scarcity.
Corrections
There are currently 113 content pages on this site, most of them dense with sourced figures. Some of those figures will be wrong, and some manufacturers will change a specification without changing a model name.
If you find an error — a spec that does not match the listing, a product that is no longer sold, a link that goes nowhere, a claim that reads as a testing claim — tell us and we will fix it. That is a genuine invitation, not a formality: contact@robotpoolcleanerpicks.com.
Written and maintained by Mike M.. More about who that is and why they write this, and how the site is funded.