Product comparison in chat: two items side by side, from catalogue data only
How a shop assistant contrasts two items using catalogue data only, shows missing values as “not stated”, and why there is no overall verdict.
In plain language
Customers often want to compare two items. The assistant then puts both side by side in a table. It only uses values from the shop’s catalogue.
If a value is missing, the table says so. The assistant does not guess. It also does not say which item is better overall.
If only one item is in the catalogue, there is no comparison. The assistant says that openly.
During the sales week, the decision is often between two offers that look alike at first glance. In the chat the question then reads: “So what is actually the difference between the two?” An assistant can answer with a paragraph full of impressions, or with a side-by-side view whose values are in the shop’s catalogue. This article shows the second form: two items in one table, with price, rating, delivery time and features, and the two product cards below it. It describes when a comparison comes about and when it does not, why a missing value stays as “not stated” and is not a difference, and why the assistant gives no overall verdict. It adds the line between this and comparative advertising under Section 6 UWG, the ban on misleading practices under Section 5 UWG in their wording, how to prepare the catalogue, and test questions for acceptance.
Labels in the illustration
- 1 · Question and query
- Visitor asks in the chat
- What is the difference
- between model A and B?
- Query: compare
- two items from your own catalogue
- only items listed as available
- in the order of the question
- Only one hit in the catalogue?
- Then one card and the question
- what to compare it with,
- no invented second item
- 2 · Direct comparison in the chat
- Criterion
- Model A
- Model B
- Price
- Rating
- Delivery
- Features
- lower
- Advantage
- higher
- average
- not stated
- longer
- shorter
- Advantage
- list in catalogue
- list in catalogue
- Model A
- Product card
- View product
- Model B
- Product card
- View product
- 3 · Rules of the comparison
- Two items required
- otherwise one card and a question
- No estimates
- missing value: not stated
- Advantage: both values
- price, rating, delivery time
- No overall verdict
- features listed, not ranked
- No overall verdict on the items: differences only from values in the catalogue
- If an item is not in the catalogue, there is no comparison, not even with a brand from elsewhere
Two offers, one question: which of the two?
Around Black Friday and Cyber Week, many people do not buy on impulse but to a plan. In the study report “Digitaler Handel in Deutschland” (digital commerce in Germany), 45 percent (Bitkom) of those who intended to use the deals around Black Friday and Cyber Week 2025 describe themselves as “Kopf-Käufer”, deliberate buyers: by their own account, they purposely buy everyday items or larger purchases when these happen to be on offer. Multiple answers were possible. What this means for the chat is our own reasoning, not a finding of the survey: anyone who buys deliberately knows what they need and may already have narrowed the choice to a few items. The last question before buying can then be a comparison question rather than a search question.
In the chat it sounds different each time: “What sets the two apart?”, “Which one arrives sooner?”, “Is the more expensive one worth it?” or simply “A or B?”. What these questions share is that the visitor means two specific items and expects an answer that looks at both at once. An assistant that replies with two separate descriptions leaves the comparison to the visitor. One that freely states “B is the better choice” takes the decision off their hands without any basis for it in the catalogue. In between lies the side-by-side view: the same criteria for both items, from the same fields, in one table. How an assistant gets to two candidates in the first place, from just a few follow-up questions, is described in the article on guided selling in chat; what happens when one of the two sells out during the sales week is covered in sold out during the sales week.
A comparison is a table, not a recommendation
The assistant places side by side the values of two items that are in the catalogue. Which difference matters for the purchase is for the visitor to decide; a verdict on which item is better overall does not belong in the answer.
What the comparison in chat shows
In XICBOT, the direct comparison is one of the displays of the product cards. The query behind it is simply called compare: it takes two items from the shop’s catalogue and hands them back to the chat window as a pair. From that the window draws a table with four rows, with the two product cards and their button below it. The buttons deliberately sit below the table and not in its cells, because a table with controls in its data cells is hard for screen reading software to interpret. Which display appears is chosen not by the wording of the answer but by the result of the query. The table shows where each row takes its value from and what happens when it is missing.
| Row | Where the value comes from | If it is missing |
|---|---|---|
| Price | From the catalogue; with a connected shop based on Shopware Community Edition, the assistant asks the shop for the price of the items shown at answering time | The cell says “not stated”; a price from memory is not provided for |
| Rating | Star average and, where the catalogue keeps it, the number of votes | The cell says “not stated”; no average taken from similar items |
| Delivery time | The item’s delivery time stored in the catalogue, in business days | The cell says “not stated”; no estimate based on the shipping method |
| Features | The item’s feature list, at most six entries, as it stands in the catalogue | The cell says “not stated”; the table derives nothing from the description text |
How meaningful a comparison turns out to be is therefore decided not in the chat but in the catalogue. A feature list that gives the weight for one item and not for the other produces a table with a gap, and no assistant can fill that gap honestly. Which fields are worth having for a shop assistant, and how dimensions, units and features are maintained so that they can be compared, is described in the article on product data for the shop assistant.
Two items or none
A comparison has conditions, and the assistant states them instead of glossing over them with wording. The first is the number: exactly two items are compared. If more come into question, the query takes the first two; further candidates need a new comparison or a recommendation row. The second condition is origin: both items must be in the shop’s catalogue and listed there as available. The third is order: if both items are identified by their reference, the one named first also appears in the first column. If a condition is missing, the display falls back rather than the answer falling away.
- Only one hit: The query returns a single card and the note that a comparison needs two items. The assistant asks what the item should be compared with, instead of making up a second one.
- No hit: If the query returns no hit, the block is dropped entirely. The assistant usually says that it has nothing on file about them.
- A model from elsewhere: If the visitor asks for a comparison with a model the shop does not carry, such as a brand from the store round the corner, the assistant does not set it against anything. It can offer to compare two items that are on file.
- Not available: An item the catalogue lists as not available appears in no card and in no comparison, not even when the visitor names it explicitly.
- Many comparisons at once: At most five tool calls are provided for per answer, and only one when the monthly allowance runs low. Anyone who wants six pairs compared in one question will not get them all in one answer.
That the assistant states a missing precondition instead of glossing over it is not a matter of style. An invented second item, or a comparison with a model for which there is no data, would be a statement without a basis. In our own measurement series, 8 of 8 questions without a basis were declined rather than guessed; what was measured were eight questions deliberately outside the material, put to our own assistant, not comparison questions in a shop. Why such a refusal is a mark of quality is explained in the article declining instead of guessing. The ceiling on tool calls matters especially during the sales week, when many conversations run at the same time; what happens when the allowance runs low is shown in the article on the monthly quota in the peak week.
Missing values: “not stated” is not a difference
The most common weakness of a comparison is not a wrong value but a missing one. If the catalogue has a delivery time for one item and none for the other, it is tempting to conclude that the second is slower or not available at all. The catalogue does not support that. An empty field only says that nothing is stored there, not how things stand with the item. In this case the table shows “not stated” in the cell and inserts no estimate. The visitor can see at a glance where the comparison holds and where it does not.
Where both values exist and differ, the table marks in three rows the value that is usually more favourable for buyers: the lower price, the higher average rating, the shorter delivery time. That is a comparison of two numbers, not a verdict on the item. If one of the two values is missing, the row stays unmarked. For features there is no marking at all, because a list of properties cannot be ordered in one direction: whether a stainless steel housing matters more than a low weight depends on the buyer. The table does not form an overall verdict on which of the two items is better.
| Situation in the catalogue | What the table shows | What the answer text should say |
|---|---|---|
| Both values present and different | Both values; for price, rating and delivery time the “Advantage” marker on the more favourable value | The difference that was asked about |
| Both values present and equal | Both values, no marker | That the items do not differ on this point |
| One value missing | The existing value and “not stated”, no marker | That nothing is stored for one item, without turning that into a drawback |
| Both values missing | “Not stated” twice | That the catalogue has nothing on this and where the visitor can look, for example on the product page |
What the assistant says in the text
The table is in front of the visitor before they have finished reading the answer text. With the result of the query, the assistant therefore receives the note not to read out all the values but to name the difference that matters for the question. Anyone who asks which item arrives sooner gets a sentence on delivery time; anyone who asks about the difference in material gets a sentence on the features. So the text does not repeat the table, it reads it for the question.
Its basis remains the same as for any other answer: the shop’s catalogue and knowledge base. In the Pro package, the XICBOT shop assistant reads the catalogue of a shop based on Shopware Community Edition; its answers are tied to the sources available to it and usually carry a source reference. If a comparison question lies outside those sources, such as “Which one lasts longer?”, it usually says that it has nothing on file about that, rather than deriving durability from price or material. The table shows how an answer without a basis differs from one drawn from the catalogue.
| Question in the chat | Answer without a basis | Answer from the catalogue |
|---|---|---|
| What is the difference between A and B? | B is the better choice overall. | Both are in the table. They differ mainly in price and delivery time; among the features only the material differs. |
| Which one is lighter? | A, it looks much more compact. | A weight is stored for A but not for B. So I cannot compare the weight. |
| Is the more expensive one worth it? | Yes, the extra cost definitely pays off. | That depends on what matters to you. According to the catalogue, the more expensive one also has splash protection; whether you need it is not something I can judge. |
| Which one is rated better? | The first one, customers like it more. | There is an average in the catalogue for A but none for B. So there is no comparison of the ratings. |
| And compared with the model from the electronics store? | That one does worse. | That model is not in this catalogue, so I will not set it against these. I am happy to compare two models that are on file. |
The rating row deserves a look of its own, because it most easily tempts an assistant into a summary: a missing average quickly becomes “only good reviews so far”. How ratings are given in the chat from the catalogue, and what competition law requires about their genuineness, is described in the article on customer reviews in chat.
Own items, no competitor: Section 6 UWG
Anyone who reads “comparison” and “advertising” in one sentence thinks of comparative advertising. The German Act against Unfair Competition (UWG) defines in Section 6(1) what falls under it:
Comparative advertising is any advertising which, directly or indirectly, identifies a competitor or the goods or services offered by a competitor.
The comparison in the chat places two items side by side that the shop itself offers. Given how the query is built, no competitor and no competitor’s offer appear in it, because it only accesses the shop’s own catalogue. Whether a competitor is nonetheless identified in an individual case, for instance when a manufacturer whose items are compared also sells directly, cannot be answered in general terms. The situation is different when the visitor expressly asks for a comparison with someone else’s offer. Then a competitor or its goods would be involved, and for such comparisons Section 6(2) sets conditions. Under point 2, anyone who advertises comparatively acts unfairly if the comparison
does not objectively relate to one or more material, relevant, verifiable and typical characteristics or to the price of those goods or services
The assistant does not draw such a comparison, because it has no data on the other item; it says so and offers a comparison within the catalogue. The idea behind point 2 still serves as a reference point for your own comparison: an objective reference to characteristics or the price rather than to impressions. Price, delivery time and stored features are details that are in the catalogue entry and can be read off there. The rating is in the catalogue too, but it is an average of customer opinions and not a characteristic of the goods. A judgement such as “sturdier” or “higher quality” is not in the catalogue as long as it is not stored there as a feature. This is exactly the line the table draws: it only places side by side what is stored in the catalogue; whether an entry is itself correct is not something it checks.
Differences with a basis: Section 5 UWG
That a comparison of two of your own items does not, as a rule, fall under Section 6 does not mean it is free of rules. The ban on misleading practices in Section 5 UWG attaches to commercial practices, and information given in a shop’s sales conversation is typically one. Subsection 2 describes when such a practice is misleading:
A commercial practice is misleading if it contains untrue statements or other statements capable of deceiving about the following circumstances
The list opens with “the essential characteristics of the goods or services such as availability, nature, execution, benefits, risks, composition”, and it also names, among other things, accessories, the date of delivery, fitness for purpose, quantity and condition. A comparison consists of exactly such statements. Anyone who claims in the chat that A arrives sooner, although no delivery time is stored for B at all, makes a statement about delivery that the catalogue does not support. The rule from the section on missing values is therefore not just a matter of accuracy. Prices are additionally subject to rules of their own, for example on the total price and on strike-through prices; these are described in the article on prices in chat.
Making the catalogue comparable
A comparison is only as good as the fields it is built from. In the same Bitkom survey, 42 percent (Bitkom) of online shoppers rate product presentation and ease of use as very important for an online shop. For the chat this means: a table in which the same details appear for both items can be read; one in which half the cells say “not stated” can hardly be read. Four points make the difference here.
- Build features the same way: What appears for one item as a weight with number and unit in the feature list should not appear only in the description text for the other. Entries named the same way make the features row readable.
- Most important first: The card takes at most six features per item. What counts for a comparison, such as size, material or output, belongs at the start of the list.
- Maintain delivery time and rating: Alongside price, these two rows carry the marker in the comparison. If they are missing for one item, the row stays unmarked and says “not stated”.
- Keep availability current: An item listed as not available drops out of every comparison. With a connected shop, the assistant asks for the availability of the items shown at answering time; the catalogue sync should still run before the sales week.
A sample is enough to begin with: place the items promoted during the sales week side by side in pairs, and read every cell saying “not stated” as a task for the catalogue. Likely places for gaps are, for instance, items that were added to the range later and whose feature list is shorter than the rest. Closing them before the sales week gives you tables that actually compare something.
Test before go-live, measure afterwards
Test questions show whether the comparison works in your own shop as described. They belong in the fixed acceptance routine and are repeated after every catalogue sync, because a sync can fill or empty fields. How such cases are checked repeatably is described in the article on acceptance testing. Five test questions cover the cases from this article.
| Test question | Expected | Not expected |
|---|---|---|
| Comparison of two available items with complete data | Table with both items, two cards below it; in the text the difference that was asked about | A recommendation on which is better overall |
| Comparison where one item has no delivery time | “Not stated” in the cell, no marker in the row | “B takes longer” or an estimated delivery time |
| Comparison for which only one item is in the catalogue | One card and the follow-up question on what to compare it with | An invented second item |
| Comparison with a model the shop does not carry | The note that nothing is on file about it, and the offer of a comparison within the catalogue | Values or impressions about the other model |
| Comparison with an item that is not available | No comparison with that item | A table containing a sold-out item |
After go-live, conversation analytics shows how often comparison questions come up. It groups the visitor questions of a period into topics, counts them and sets each topic against the count of the preceding period of the same length. Topics with lead terms such as “difference”, “compare” or “better” suggest that visitors in the chat are torn between two items. The analysis knows no word meaning, only matching word parts; related questions without shared word parts need to be added up when reading. If a comparison question ends without a source reference, a line usually appears among the knowledge gaps, and it leads to an entry in the catalogue, not a new wording in the system instruction.
Limits: what the comparison does not do
- It does not set more than two items against each other in one table; for more candidates there is the recommendation row.
- It does not compare items that are not in the shop’s catalogue, nor any listed there as not available.
- It does not estimate missing values; what is neither in the catalogue nor in the knowledge base it is not meant to claim.
- It does not crown an overall winner. Which difference matters for the purchase is for the visitor to decide.
- It promises no effect on conversion rate or returns. Whether a comparison makes the decision easier depends on the range and the catalogue data.
- It does not judge whether a comparison or a wording is legally permissible.
Frequent questions
How many items can the assistant compare at once?
Two. The direct comparison places exactly two items side by side in one table. For more candidates the assistant usually shows a recommendation row with several cards; the visitor can start a further comparison with a new question.
What happens if a value is missing for an item?
Then the cell says “not stated” and the row receives no marker. The missing value is not estimated and is usually not presented as a drawback of the item either, because it only says that nothing is stored in the catalogue.
Does the assistant say which item is better?
An overall verdict is not provided for. Where both values exist, the table marks the more favourable value for price, rating and delivery time. Which difference matters for the purchase is, as a rule, for the visitor to decide.
Can the assistant compare an item with another shop’s offer?
Not as the query is built: the comparison only accesses the shop’s own catalogue. If a visitor asks about a model that is not in it, the assistant usually says that it has nothing on file about it and offers a comparison of two items that are.
Is a comparison in chat comparative advertising within the meaning of Section 6 UWG?
According to the wording of Section 6(1) UWG, it concerns advertising that identifies a competitor or its offers. A comparison of two of your own items generally does not do that. Whether a competitor is nonetheless identified in an individual case is a matter for legal advice.
Where do price and availability in the comparison come from?
From the shop’s catalogue. With a connected shop based on Shopware Community Edition, the assistant asks the shop for the price and availability of the items shown at answering time. An item listed as not available usually appears in no comparison.
Ask it yourself, Assistant of this website
Automated assistant