A Hundred Service Pages, Four Answers to “What Do I Get?”

The page underneath the price

VIT MARKET, the services arm of the VITON13 platform, publishes 100 listings at viton13.com/vit-market/services. Each one is stored as a record with thirteen fields: a number, a title, a difficulty word, a category label, a description, two price strings, a delivery estimate, a slug, a group key, a numeric price floor, a numeric price ceiling and a true-or-false featured flag. That is the entire published dataset behind a listing.

Everything else a visitor meets on the page is assembled at render time out of those thirteen fields. The bullet list of what the service delivers is generated. The three questions and answers underneath it are generated. The block of related services at the bottom is generated. None of it is typed by a person for that particular service.

This matters because of how anyone actually buys. You narrow a catalogue to two candidates, open both pages side by side, and hunt for the difference. The instinct is right. What follows is a measurement of how much of that page is capable of showing a difference at all — recomputed from data/vit-market-services.js and from lib/vit-market-seo.js, the module that builds the questions and the machine-readable markup.

Four answers to the question “what do I get?”

Every listing page shows a four-bullet list of deliverables. The first bullet is chosen by the listing's group; the other three are constants. Across the whole catalogue that produces exactly four distinct lists — one for each of the four groups — and each list is shown on exactly 25 pages. Four hundred bullets are rendered in total, drawn from seven distinct strings.

The three that appear on all 100 pages are these: “A concise brief and an agreed scope before work begins”, “An interim review with a clear project status”, and “Final testing, source handoff and practical instructions”. The four group bullets add responsive implementation and SEO markup for websites, a prioritized plan and a before-and-after report for search work, a process map and error handling for automation, and response-quality evaluation, cost limits and a safe shutdown procedure for AI.

The same four bullets are also joined with semicolons and published as the serviceOutput property of the Service markup embedded in each page, which means the machine-readable statement of what 25 different services output is byte-for-byte identical. A search engine reading those 25 pages is told the same thing 25 times.

None of this is dishonest, and it should not be read as sloppiness. A promise about method is a real promise, and applying it uniformly across a family of jobs is a defensible way to run a shop. It simply is not an answer to the question a buyer comparing two listings is asking, because the answer does not move when the listing does.

Three hundred questions written by two sentences

Each page also carries three questions and answers, so the catalogue publishes 300 of them. The first asks how much the named service costs, the second asks how long it takes, and the third is identical on all 100 pages: “What is included in the service?”

The first answer quotes the stored floor and ceiling and then adds a fixed sentence of fifteen words: the final estimate is confirmed after the project scope, constraints and delivery format are reviewed. The second answer quotes the delivery estimate and adds a fixed sentence of fourteen words about timing being confirmed before work starts. The third answer is the four deliverable bullets joined into a paragraph.

That fixed fifteen-word sentence is worth stopping on, because it is the catalogue telling you plainly what it is doing. It names three inputs — scope, constraints, delivery format — and states that the number becomes an estimate only once they exist. On all 100 pages, and inside the machine-readable question markup that assistants and search engines quote back, the published price is declared to be a pre-brief number.

The related block that has only ever recommended sixteen services

At the foot of every listing sit three related services. The rule that picks them is simple: take the other members of the same group and keep the first three in file order. It is not a similarity match on price, difficulty, category or subject.

Run it across the catalogue and the arithmetic is stark. There are 300 related-service slots. Sixteen distinct services ever occupy one. Eighty-four listings out of a hundred have never appeared as a recommendation anywhere on the site. Twelve services appear 24 times each and four appear 3 times each, which accounts for all 300 slots.

Concretely: 88 of the 100 pages show one of only four triples, because 22 of the 25 members of each group see that group's first three entries. If you want the listings that are genuinely comparable to the one you are reading, the group filter on the index page is the tool; the block at the bottom of the page is a shelf's front row.

Forty-five words that are yours, seventy that are everyone's

Put a number on it. Counting the English edition of the catalogue, and treating the title and description as the listing's own prose, the average listing contributes 45.07 words of its own. Counting the four deliverable bullets, the fixed sentence in each of the first two answers and the third question, the average page carries 70.25 words that at least 24 other pages carry verbatim.

That puts the listing-specific share of the compared prose at 39.08 per cent. The thinnest listing has 29 words of its own — the search-cannibalisation entry — and the two fullest have 63, the GEO and AEO listing and the weekly-report listing. No listing anywhere in the catalogue writes more of its own page than the template writes for it.

One caution on this figure specifically: word counts are a property of a language edition, not of the underlying record. The Russian and Spanish editions of the same 100 listings are worded differently and would produce different totals. The 45.07, the 70.25 and the 39.08 per cent describe the English catalogue text and nothing else.

Two pairs where every published number matches

There are 4,950 ways to pick two listings out of a hundred. Nineteen of those pairs sit in the same group, carry the same difficulty word and share an identical price floor and ceiling. Two of the nineteen also share their delivery estimate word for word — and those two pairs are the clean experiment this whole exercise was looking for.

The first is “AI agent for cold sales” and “AI analyst answering database questions”. Both are on the AI shelf, both are labelled Advanced, both are published at $1,200 – $3,500, both quote 14–25 working days, and neither carries the featured flag. Their deliverables are identical. All three of their answers are identical, character for character, because the answers are built from the price fields and the delivery estimate and both listings hold the same values. Their related-services blocks are identical too. What differs is the title, the description and the category label — Sales against BI and data.

The second pair is “AI screening of resumes for a specific vacancy” and “Personalization of letters for each lead”, both Experienced, both $600 – $1,800, both 5–10 working days. Everything said above applies, with one extra difference: the printed price label. One reads as a plain band; the other adds a qualifier saying the money buys the pipeline.

What the templated answer throws away

That qualifier is not an isolated case. Twenty of the catalogue's 100 dollar labels carry something beyond the two numbers: a unit the price is charged against, such as per letter, per page, per video or per card; a stage the price covers, such as setup, diagnostics or the pipeline; or a comparison with what the same job fetches elsewhere.

The generated answer to “how much does this cost” is built only from the stored numeric floor and ceiling, so on every one of those twenty pages it drops the qualifier. The page's own answer is less informative than the label printed above it, and the gap is not decorative — a band that buys a pipeline and an identical band that buys a finished job are two different offers.

The practical rule is short. On a listing page, read the label and ignore the answer. On the two pairs above, the answers were never going to separate anything: they are functions of numbers that already matched.

What a listing page cannot tell you

The catalogue publishes no field for revisions, no ownership terms, no support window, no response time and no remedy if the result does not work. Absence in a dataset is not absence in practice — a real engagement may settle every one of those in the first conversation. What the data supports is only the narrower claim: you cannot learn them from the page.

Nor can the page tell you which of two matched listings is harder for your particular situation. A category label is a filing decision, not a scope. Knowing that one listing is filed under Sales and the other under BI and data is real information about how the shop is organised and no information at all about which one will take longer against your systems.

The pairing test above followed the stored numeric floor and ceiling rather than the printed label, which is why the second pair counts as a match on price despite the difference in wording. Anyone re-running it against the printed labels would find one pair, not two. Both readings are defensible; the method is stated so the choice is visible.

Five questions that pull two listings apart

First: which of the two published numbers would appear on the invoice, and does the label carry a unit or a stage after them? Second: if there is a unit, what is the quantity — how many pages, cards, sources, locations or documents — and who counts it? A price per card and a number of cards are one decision, not two.

Third: what is fixed at signature and what is agreed afterwards? The catalogue's own bullet says the brief and the agreed scope come before work begins, which is a different moment from before the price is published. Fourth: who owns the result and the accounts it runs on when the last invoice is paid. Fifth: write down what “done” looks like in one sentence you could actually test, and see whether both sellers will sign the same sentence.

Two listings that agree on group, difficulty, price and schedule are not a coincidence to be resolved by staring at the page harder. They are the catalogue telling you it has run out of things to say, and handing the rest of the decision to the brief you have not written yet.

Decision framework: how to compare service listings

Compare service listings by evidence and acceptance, not by the nouns in their titles. A useful card distinguishes the initial state, final artifact, measurable standard, client input, exclusion and change mechanism.

Repeated promises such as professional quality or fast delivery create no comparison surface. Specific constraints may sound less impressive but reduce the uncertainty a buyer is actually purchasing.

Hide provider names, extract six fields into a blind table and mark every unsupported promise; restore the names only after deciding which missing answer would change the purchase.