AI for marketplace sellers: why a listing gets impressions but no purchases

AI for marketplace sellers: why a listing gets impressions but no purchases

Impressions rise, purchases do not — and the first thing a seller does is spend on promotion. The money goes nowhere: people see the listing and do not buy, and advertising multiplies that problem. Below is the funnel taken apart by stage, with the causes in the order to check them.

Impressions rise, purchases do not — and the first thing a seller does is spend on promotion. The money goes nowhere: people see the listing and do not buy, and advertising multiplies that problem. Below is the funnel taken apart by stage, with the causes in the order to check them.

There are impressions and no sales

The most common and most expensive situation on a marketplace.

The listing shows in search and the impression figures are respectable. Purchases: a handful. The seller draws the only conclusion available: not enough traffic, must invest in promotion.

He invests. Impressions rise, sales barely move. The money has gone nowhere, and worse, gone unnoticed, because rising impressions create a sense of movement.

The cause is elsewhere: people see the listing and do not buy. Advertising does not cure that; it multiplies it — every new impression runs into the same bottleneck.

There are not many such places, they are predictable, and they are fixed in order. Let me take it apart on a real assistant's run.

Who analyses the product card

I took the Marketplace content master (the Ozon one) from the catalogue — not "a model in general" but a system for a specific platform.

The marketplace content master — its specialisation and range of tasks

What it covers by its own description: normalising data and choosing a category, SEO optimisation and copywriting, briefs for visuals and infographics, validation against moderation rules, and forming hypotheses for A/B tests.

Note the fourth point. Checking text against platform limits, banned words, unverified claims and legal risks — the things that get listings sent back for revision, and the things a general model without specialisation does not know.

The funnel taken apart

The task was set with no data at all, in one sentence:

Analyse why a listing gets impressions but no purchases. Name the typical causes in the order to check them and what is done about each.

The answer is built strictly along the platform funnel: impressions → clicks → basket → purchase. The diagnosis proceeds in steps, and that matters: you fix the stage that sagged rather than everything at once.

Stage 1. Impressions but no clicks

The first funnel stage — four causes of low CTR and what to do about each

Somebody sees the product in the results and does not click. Four causes, each with a ready action:

A weak main image. It blends into the results, a poor angle, overloaded infographics. Action: a high-contrast photo, the product large, one or two key advantages placed on the image.

An uncompetitive price or no visible discount. Action: monitor prices in the category, add a discount flag.

A low rating or few reviews. A rating below 4.5, or noticeably fewer reviews than neighbours. Action: enable points for reviews, deal with negatives, lower the price while the first ten to twenty ratings accumulate.

Off-target impressions. The listing shows for queries that are relevant in meaning but not to the buyer. Action: check the category and clear the keywords of anything too broad or junk.

The last point explains the most galling case: there are many impressions because the listing is catching the wrong people. Advertising here makes it worse, not better.

Stage 2. They click but do not add to basket

The second and third stages — from basket to payment

This is where most is lost, and where people look least often.

The gallery does not answer the questions. No dimensions, materials, details or use case. Action: six to twelve slides that successively set out the benefits, close the fears and show the product in use.

Missing specifications. The buyer cannot find a critical parameter — compatibility, exact size, weight — and leaves. Action: fill in every mandatory and additional attribute, and remove discrepancies between text and photos.

A weak description. Overloaded with keywords, unstructured, answering nothing. Action: rewrite it on the "audience → benefits → scenarios → contents" scheme and add a block of ten to twenty frequent questions.

Stage 3. Added to basket and not paid for

Slow delivery — the item sits in the seller's remote warehouse. Action: move it to the platform's warehouse and distribute stock across high-demand regions.

Unclear contents or warranty. The customer doubts what will actually arrive. Action: contents as a separate block in the text, plus a text overlay with the warranty on one of the slides.

What matters more than the analysis itself

The order. The analysis follows funnel stages rather than a list of tips. That lets you avoid fixing everything at once: you look at where it sags and work only there. The saving is usually severalfold.

An action next to every cause. Every problem ends with what is done about it. That is the difference between "your listing is bad" and a working document.

Reviews as a source rather than a problem. The most underrated move: the questions under a product show what the listing is missing. If the same question has been asked twenty times, the answer to it belongs on the infographic — and every question asked means a dozen people who did not ask and left.

Two things people usually do not know

The assistant has checking modes. The answer ends with an offer to assess the work and suggest improvements. There is a focus group mode — the model goes through its own answer with a buyer's eyes — and refinement against noted criteria. The first draft becomes a checked document without your involvement.

Assistants are put in a chain. The output of one goes into the input of the next: this one assembled the infographic brief, the prompt generator turned it into an image prompt, the model generated the slides. What works is a conveyor rather than an individual answer.

How a marketplace seller earns from this

First — your own goods. Analysing and rebuilding the listing before launching advertising. That order specifically: advertising multiplies what is already there, and money must not be put into an unfixed listing.

Second — as a service to other sellers. Relaunching listings end to end: funnel analysis, images, infographics, description, a procedure for handling reviews. From our bot's feed of 5 986 jobs over four months, marketplace listings run $245–730 for a set — against a median one-off job of around $60. That is a different unit of measurement: not a picture but a result.

And a separate market observation: in my audience research, marketplaces came last for interest. Fewer people want to work there than is generally assumed. For anyone already inside, that is good news — competition among contractors is lower than in other niches.

Where to start

The "E-com" category holds thirteen assistants: for the major marketplaces, plus a visual merchandiser, a purchasing expert, a shop manager. In total the platform holds 200+ AI assistants across 14 categories.

Image generation for infographics is paid from the wallet on usage — the cost is visible before you press, and on a failure the money is returned.

Registration is free and opens three days of full Basic access — all the assistants, the prompt texts, the skill files and six Basic mini-courses.

Do one thing today: open the questions on your own product and count how many times the most frequent one recurs. The answer to it belongs on the infographic. That is the cheapest fix with the most visible effect — and it needs not a penny of advertising budget.

The second approach: not a chat but a conveyor

Everything above is work in dialogue with an assistant. There is another format, and for volume it suits better.

Cascading prompts in spreadsheets

The course “Earning with ready templates” contains spreadsheets with an integrated model: a large task broken into micro-steps, each living in its own cell and taking the previous result as input. The formula can be dragged down a thousand rows — and a thousand tasks are processed at once. How this works step by step is covered separately: Thirty posts in one launch.

What is available for a seller:

Spreadsheets like these (the lessons "How to work with GPT Sheets" and "Content factory") suit listings too: descriptions, specifications and texts across the whole catalogue at once. Drag the formula down a thousand rows and a thousand items are processed.

The course is included in the Basic plan.

When you need images, video and sound

A spreadsheet is enough for text. If a chain needs images, video or sound, hand the task to Cerebrum: describe it in plain words, the right Gen AI models are picked automatically, you pay from the balance by actual use, and the price is visible before launch. Single generations live in the Gen AI section.

Thirteen assistants for e-commerce

The "E-com" category in the catalogue, Basic plan — by platform and by task:

A set of listings without a photo studio

The main image, angles, lifestyle shots and infographics with callouts for the major marketplaces are made by the models in the Gen AI section — the 3:4 format is set at generation. The easiest way is to hand it to Cerebrum: describe the task and give a link to the product photo, and the models are picked for you. The same route produces a vertical clip "as shot by an ordinary customer", a cinematic 5–15 second ad and 6–9 creatives for A/B testing. On the Full plan the AI SKILLS connector for Claude does the same.

The price is visible before launch and money for anything not delivered returns to the balance. Wallet payment from $5.

A ready solution for this profession

The skill “Product-card photos for a marketplace” is a direct hit on a seller's task. From 1–3 product photos and 3–5 benefits it builds a set of 6–8 frames: the main photo, angles, lifestyle shots and infographics with callouts, in 3:4 for Ozon and Wildberries. It works through the AI SKILLS connector for Claude (Full plan); the skill file comes with the Basic plan.

For clothing there is the skill “Virtual try-on for product cards”. And in the Gen AI section there is the “Product Shot” model: a studio photo of the product on a new background from a scene description.

What this changes in the economics. A set of listings costs $245–730 on the market, while preparing one item by hand takes hours.

What to pick up on the platform for this task

What to read next

AI for designers — how to assemble a series of listings in one style.

A 3D model from a product photo — what to add to a listing when photos are not enough.

A clip from a product photo — eight advertising formats without filming.

The AI freelancer price list — what a set of listings costs on the market.

AI video generation in 2026 — the full guide to video models.

Also on AI SKILLS