Comparison
Refik and Zowie, read on the same day.
One calls itself an AI agent platform for enterprise customer service. The other is a shopping assistant that only exists inside one Shopify storefront, and every figure it charges is printed on a page in this site. The rows below are honest about one thing before they are anything else: this is not a price against a price, because on the day we read it only one side’s prices were readable at all.
Method
What was read, how the zeros were checked, and what a blank cell means.
The right-hand side of every row was read out of Zowie’s own public pages on 10 October 2026, with a plain command-line fetch that followed redirects. No analyst report, no review site, no second-hand summary of either. Where a cell carries a sentence in quotation marks, that sentence is theirs word for word, and the addresses it came from are printed above the table. The left-hand side was read out of our own source code on 4 October 2026, and the plan figures come from the pricing page in this same directory — a page whose build fails if a figure in the code changes without the page changing with it.
Two rows below rest on a count that came out at nothing: a word that did not appear on the pages read. A count of nothing is worth nothing unless you know the reader was awake, so the same reading ran two controls. It looked for each brand’s own name on every page it fetched and required at least one hit. It also looked for a sixteen-character string generated at the moment of reading, a string that exists nowhere, and required no hits at all. Both controls held across the whole sweep, which is what makes these particular zeros measurements instead of silence. The control string is deliberately not printed here, because writing it down is how you poison the next reading that looks for it.
A blank cell is a third state, and it is on purpose. It does not mean “they do not have it”. It means the figure was not readable on the day and the pages we read, so putting a dash or a nought there would turn a limit in our reading into a claim about their product. Every blank cell is named underneath the table, with the reason it is blank.
Three things are missing from this page by design. There is no verdict row and no score: a published figure and an unmeasured one cannot be set against each other, and any number that came out of trying would look like a measurement without being one. There is nothing about how well either assistant answers a shopper, on either side, because nobody has put the same questions to both against a real catalogue. And there is no sentence anywhere below about why either company sells the way it does, because motive is not something a page fetch can measure.
The rows
Nine rows, and the four that only one side could fill.
Read the first row before the rest, because it is the reason four cells on the right are blank. The figures that would go in them were not on the addresses read, so those rows carry one side each. A row with one side filled is a record of what we found; calling it a comparison would be the first dishonest sentence on this page, and it is not going to be written here.
Zowie read from getzowie.com, the integration page that getzowie.com/integrations redirects to, getzowie.com/pricing and getzowie.com/sitemap.xml; Refik read from this repository and from the pricing page in this directory.
| What the row measures | Refik | Zowie |
|---|---|---|
| What it costs, and where the figure sits | ||
| Where the price is published | A page in this directory, open to anyone who opens it: three plans, both billing cycles, and the conversation allowance each plan includes. Nothing stands between a reader and the figure. | getzowie.com/pricing answered HTTP 404 on the day it was read, and the site’s own sitemap.xml — which answered normally and listed hundreds of addresses — carried no pricing address among them. Two signals, taken separately. On the three pages read that day the word “Pricing” appeared no times, while the word “demo” appeared between two and four times on those same pages. |
| The list prices, in USD | Starter, $39 a month or $390 a year. Then $99 a month or $990 a year. Then $249 a month or $2,490 a year. Each figure is for the whole shop rather than for each person who answers. | |
| What the subscription is counted by | Conversations inside a billing period, in three bands: 300, then 1,000, then 3,000. What counts as one conversation is defined once, on the pricing page, rather than restated here where a second definition could drift from the first. | |
| Charges on top of the subscription | On monthly billing, $0.15, $0.12 or $0.10 for a conversation past the band, and it stops at one band’s worth rather than running on. On yearly billing, none at all. | |
| Free trial | 14 days on every plan, with a conversation allowance scaled to the plan it belongs to. After the trial there is no free tier to fall back to. | |
| Who it says it is for, and where it runs | ||
| How each side describes itself | A shopping assistant on the storefront. It recommends out of one shop’s catalogue, looks up an order, and hands a question it will not guess at to a person. | The home page title reads “AI Agent Platform for Enterprise Customer Service - Zowie”. |
| The buyer each side names | One Shopify store, often with one person answering behind it. That is why the price is per shop and why the trial starts without anybody having to talk to us. | The home page headline reads “AI you can hand your customer to The AI agent platform leading enterprises run in production. Built for the brands where getting customer-facing AI wrong isn’t an option.” |
| Platforms named on the pages read | Shopify, and nothing else. The catalogue, the policies and the order lookups all come out of one store’s Admin API, and there is no second commerce platform anywhere in the code. | On the three pages read that day the word “Shopify” appeared no times. Asking for getzowie.com/integrations ended at getzowie.com/agent-connect, whose page title reads “Agent Connect - AI Agent Integration Platform - Zowie”. |
| Search standing and traffic | ||
- The four blank cells on their side have one cause between them. The tiers and the amounts are recorded as not measured in the fact base this page is built from. We asked for the pricing address and got HTTP 404; we read the site’s own sitemap and found no pricing address listed in it. That is a measurement of two addresses on one day. It is not a finding that the product has no price, and it is not a finding about how the company sells — neither of those was measured, so neither is written anywhere above.
- Search standing and traffic, both cells blank. The tools that would measure search visibility were not connected on the day the fact base was built, and they were not connected for our own side either. Nothing was estimated to fill the gap, so the row stays blank on both sides rather than carrying a figure nobody checked.
- One of their own claims is a percentage figure, and it is named here rather than reproduced. Their home page carries a percentage claim about how workflows execute. This site prints no percentage figures at all, ours or anybody’s, because a figure lifted out of the page that defines it starts to read like our own measurement. So the claim is recorded as existing and the number is left where it was published.
- Which three pages the word counts cover is not named. The fact base records that three pages were read and counts words across them; it does not label which three, so this page does not label them either. The addresses that were read are listed above the table, and the count stands over three of them without our being able to say which three.
- Neither side’s answer quality is here. Not theirs and not ours. Measuring it would need the same shoppers putting the same questions to both, against a real catalogue, and that has not been done on either side.
The difference that matters
A published figure and an unmeasured one are not two prices.
A published figure lets somebody decide alone. A shop owner can read three numbers, pick one, start a trial and never speak to us at all; if the answer is no, the no costs an afternoon. An unmeasured figure is different in kind rather than in size. The reader of this page cannot put it next to ours, cannot work out which is larger, and cannot tell from the table whether it is larger for a small shop and smaller for a large one. That is the honest shape of the first half of this comparison, and the table shows it as four one-sided rows rather than hiding it inside a sentence.
Our own page does the same thing above a point, and a page like this one has no business pretending otherwise. Past the top band there is no fourth card on our pricing page, and the enterprise page in this site gives the reason in its own words: a card implies a price, and a price we would have to publish before knowing what we were agreeing to. So the honest version of the first row is not “one side publishes and the other does not”. It is: one side publishes three figures and then stops, and on the pages we read the other side published none of the three. Where our own figure stops being published, that is written on the page where it stops.
The second difference is in the words each side picked for itself, and those are quoted in the table rather than described. Their title says platform, enterprise, customer service. The address that asks for integrations ends at a page whose own title says integration platform. Something that sits between many systems is a different shape of product from something that sits inside one, and a different shape of product usually has a different shape of price. Ours is narrow by construction: one commerce platform, one storefront, one channel. The web storefront is the only place a shopper can reach it — messaging apps, email and SMS are not built, and that is written on the integrations page rather than left to be discovered after the first invoice.
There is one more limit worth putting in plain words here, because it is the kind of thing a shop finds out late. The server can open a ticket in one help desk. That help desk is not named on this page, because which help desk a shop already runs was not part of what we read on 10 October and we are not going to guess at it. The settings screen for choosing a handover target is not built either. So a shop that wants the storefront to hand its hard questions into an existing queue cannot wire that up from our panel today, whatever queue it is.
Who each one fits
Where we would send you to read their page instead of ours.
Go and read their page if your organisation buys software the way large organisations buy it: a requirement list, a security review, several customer-facing systems that have to agree with one another, and a process that expects a quote rather than a card charged on the spot. Their own headline names that buyer in their own words, quoted above, and a sentence about the kind of company a product is built for is the only sort of claim we can repeat, because they wrote it and we dated it. If that description fits you, the figures in our column are the wrong figures anyway: they are a price for one shop.
We fit you if you run one Shopify store and the questions that cost you money arrive before the order exists. Which of these two suits me. Is this in stock in my size. Where is the order I placed on Friday. Can I send this back. Those arrive on a product page, answering them needs one shop’s catalogue and one shop’s policies rather than a queue, and you can find out what it costs by reading a page and then paying for a single month.
Neither of us fits you if you are not on Shopify. We do not run anywhere else, and there is no second commerce platform sitting in the code waiting to be switched on. Neither fits either if what you actually need is a person. Both sides put a human at the end of the path; ours does it by opening a handover and telling the shopper somebody will pick it up, and it is instructed never to tell a shopper that a refund has been approved. If there is nobody to pick the handover up, no assistant on either side of this table repairs that, and no price on either side of it is the problem.