ChatGPT recommends my competitor: what being named second actually costs

You have already done the checking. You asked the question a buyer would actually type, and you are in there. Sometimes. What is also in there, higher up and in more of the answers, is the business down the road.

That is a different problem from being invisible, and it has a different answer. If you have not measured anything yet, start with the manual check for whether AI assistants recommend your business. It costs nothing and it takes about half an hour. This note picks up where that one ends, at the result nobody writes about: named, but second.

We build a paid AI visibility monitor for Wix sites, so read the rest accordingly. The part we would want to read first is the arithmetic, so that is what is below, in full, including the weights we chose and the reasons you might disagree with them.

Being outranked is three separate problems

When an owner says a competitor is beating them in AI answers, they almost always mean one of three things. They are worth separating, because they cost different amounts and they have different fixes.

1. How often you are named.

   Ten buyer questions, they appear in nine, you appear in three. That is a coverage problem: for seven of those questions, nothing about you reads as an answer to that question. This is the one people notice, and it is the one that carries the most weight.

2. Where in the answer you are named.

   You are both there. They are the first sentence with a reason attached, you are fourth in a list at the bottom. Buyers read these lists from the top and stop early, the same way they always did with search results.

3. How much of the answer is somebody else.

   You are named once and five other businesses are named alongside you. Being one of six is not the same as being one of two, even when your mention rate has not moved at all.

These move independently, which is the part that trips people up. A store can hold a perfect mention rate for a quarter and still lose ground the whole time, because every answer it appears in now names four more shops.

What it costs, in the arithmetic we actually use

Here is our scoring model. Not a simplified version of it for a blog post: this is the calculation that produces the number in the product.

100 * (0.60 * mention rate + 0.25 * prominence + 0.15 * share of voice)

The three terms are the three problems above:

• Mention rate. Answers that named you, divided by answers asked. Named in three of ten is 0.3.
• Prominence. The average of a position weight across the answers where you were named. First counts 1.0, second 0.7, third 0.5, fourth or lower 0.3. An answer where you are not named at all counts 0 and does not drag the average, because it has already been counted once in the mention rate.
• Share of voice. Your mentions divided by every business name across the whole set of answers. This is the term that notices a crowd.

Two things fall out of that which are worth knowing before you spend a weekend on anything.

Position is expensive. A fourth mention counts 0.3 against a first at 1.0, less than a third. Getting onto the list is not the win, and a tool that reports only whether you were mentioned is hiding most of the movement from you.

A competitor joining an answer below you costs almost nothing. Share of voice is 15 of the 100 points, and it is the only term that moves when a business is added to an answer you were already in without displacing you. What costs you is being pushed down, not being joined.

Take one assistant and ten buyer questions. You are named in five of them, first every time, and every answer in the set names four businesses in total. Then hold everything about your business still and change only who else is in the answers.

Same business in all four rows. Same mention rate in all four rows. Nearly 18 points of a hundred between the top and the bottom, and not one of those rows involves you losing a mention.

The middle two rows are the useful pair. The same competitor, in the same answers, costs you about a third of a point when it lands below you and nearly 8 points when it lands above you. If you are going to be annoyed about one of them, be annoyed about the right one.

Now argue with it

Those three weights are a judgment, not a law. We chose 0.60 for mention rate because being named at all has to happen before anything else matters, 0.25 for position because buyers read from the top but do keep reading, and 0.15 for share of voice because a crowded answer costs you something real but less than a demotion does.

An assistant does not publish weights. It does not have any. There is no ranking table behind an AI answer to be measured against, so every score in this category, ours and everyone else’s, is a convention for turning what was observed into one number you can track. Ours is on this page so that you can argue with it. If you think position deserves more than a quarter of the score, you are not wrong in any way that can be proven. You have a different model, and now you can say exactly how it differs from the one your number came out of.

Why the assistant named them and not you

There is no grudge, and there is no list of approved businesses. When a question sounds local or commercial, assistants generally search the live web, read a handful of pages, and summarise what they found. A name in an answer is downstream of three plain properties.

• It could be retrieved. The pages were reachable by the crawlers the assistants use for search, and the site was indexed. If the search crawlers are blocked, nothing else on this page applies.
• It could be parsed. The facts a recommendation needs, what the business does, where, for whom, at roughly what price, are written in sentences on a page rather than implied by photographs or trapped in an image.
• It was corroborated. Somewhere other than their own website says the same thing about them. Assistants are careful about recommending claims that only the claimant makes.

And the possibility nobody enjoys: sometimes the competitor is the better answer to that specific question. They say plainly that they do the thing. You may well do the thing too, and never have written it anywhere. The hub post ranks these factors by how much control you have over each one, and that ranking has not changed.

One caution before you build a theory. Assistants personalise and they vary day to day, so “my competitor is always first” is very often “my competitor was first in the four answers I happened to look at”. The size of that variation is the last section of this post, and it is larger than most people assume.

What actually moves it

You do not move up by pushing anyone down. You move up by becoming a better answer to one specific question. In rough order of what we see move a score:

1. Write the page that answers the question you keep losing.

   Take the two or three questions where a competitor is named and you never are. Write a page or a section that answers each one directly, in the words the question used. This is the highest value change on the list, the one that keeps working, and the one most people skip because it is writing rather than a setting.

2. State the deciding facts as text.

   Service area, hours, price range, what you specialise in, and what you do not do. A stated fact can be repeated by an assistant. An implied one has to be guessed at, and guessing is the thing they avoid.

3. Add structured data for what you are.

   LocalBusiness for a service business, Product for a store. It does not force anyone to cite you. It removes ambiguity, and ambiguity is what leaves you off a shortlist of three.

4. Confirm the AI search crawlers are not blocked.

   Wix has a setting for this and some owners have turned it on deliberately. If yours is on and you are wondering why you are never named, you have found your answer and you should decide on purpose rather than by accident.

5. Check your llms.txt says something specific about you.

   Wix generates one for your site. The note on llms.txt for Wix sites covers where to find it and what belongs in it. If the summary reads as generic, that is what a machine reading your site concluded about you.

6. Fill in the product SEO fields you skipped.

   Empty fields are common on Wix stores because the upload flow does not insist on them. Filling them is dull and close to free.

7. Get described somewhere that is not your site.

   Directories, review platforms, local listings, a roundup you are actually eligible for. This is the slowest item here and the one your winning competitor most often has.

Since we are asking you to trust a fix list, here is exactly how ours divides. Of the eight fixes our checklist raises, five are applied to your site from the dashboard with a preview first and one click undo: llms.txt, crawler access in robots.txt, LocalBusiness structured data, Product structured data, and empty product SEO fields. We guide the other three and deliberately do not make them for you: FAQ content that answers a buyer question, Bing IndexNow registration, and page titles and descriptions. The integration page is the full record of what gets read, what gets written, and what we will not touch.

What will not change it, at all

This section matters more than the last one, because the market around this product is full of the opposite.

• You cannot demote a competitor. There is no submission form, no appeal, no setting, and no supplier with a lever for it. The only half of the comparison you control is your own half.
• Nobody can promise you a place in an answer. Not an agency, not a tool, ours included. The answer is regenerated per question, per assistant, per day. Anyone selling a position is selling something they do not have to sell.
• Naming your competitor across your own pages transfers nothing. A comparison page you cannot support is a liability rather than a lever, because assistants quote pages, and the claim you cannot back is the one that gets quoted.
• Volume is not corroboration. Twenty thin pages written in an afternoon are still one source saying one thing about itself. The signal that helps is other people describing you consistently.
• A file is not a promise. llms.txt is a proposed convention and no provider has committed to reading it. It is cheap and worth doing. It is not a mechanism.

We would rather say this on our own marketing page than have you learn it after a month of paying us. There is no lever in our product that moves a competitor down, and there is no version of this product where there could be.

Do not read noise as a result

Before you act on a difference between you and a competitor, or between this week and last, you need to know how big a difference has to be to mean anything.

We ran three consecutive weekly scans against a site we did not change at all. The scores were 46, then 53, then 58. Nothing about the business moved. Assistants personalise, they search a live web that changes underneath them, models are updated without announcement, and a question set is a sample rather than a census.

So every score we report comes with a stability band: the amount of movement that would be ordinary for that particular scan. The band is arithmetic, and it is this:

flip = 60 / (questions * assistants)
band = round(1.5 * flip + 0.15 * spread), clamped to between 2 and 15

A flip is the smallest real event we can observe: one assistant changing its mind about one question. Mention rate carries 60 of the 100 points and the headline is the average across the assistants, so a single flip moves the score by 60 divided by questions times assistants. We treat about one and a half of those as ordinary drift. Spread is how far apart the assistants scored you, and it widens the band a little, because a week’s average is less settled when the assistants disagree with each other.

What that means in practice: a five question scan across three assistants lands on a band of about six or seven points. A twenty five question scan lands on two or three. So a four point rise on a five question scan is not an improvement, whatever you changed that week. The same four points on a 25 question scan is a real move. Asking more questions does not make the assistants more consistent. It makes your sample large enough that their inconsistency averages out, which is the actual reason a wider question set is worth anything.

The same test applies to your competitor. “They were named more often than me this week” across five questions is inside the noise as well.

Two honest notes about the band. It is presentation, not measurement: it is read time arithmetic over scores that were already stored, so we can retune it without altering a single historical number, and we would rather retune it than defend it. And Wix says the same thing in its own words about the free AI Visibility Overview in Wix Analytics, warning that running the same questions again may produce different results at different times. That warning is correct and fair. The thing it does not come with is the size of the swing, which is why we publish ours. We compared the two tools properly here, including the cases where the free one is the whole answer.

Watching the gap instead of the number

Everything above can be done by hand, and the method in the hub post is the way to do it. The part that does not survive contact with a real month is the repetition: the same questions, the same assistants, the same spreadsheet, every fortnight, while running a business.

That is the gap Patesco fills. It reads your published Wix site, writes the buyer questions, asks ChatGPT, Gemini and Perplexity on a weekly schedule, scores the result with the formula at the top of this page, names the competitors that keep appearing next to you, and emails you what moved and whether it cleared the band. The free plan runs 5 questions across the three assistants and shows the real result, competitor names included. Premium runs 25 questions, which is the plan that narrows the band enough to read small movements, at $19 a month.

If the manual version is enough for your site, use the manual version. It is the same idea, and the formula is right there.

FAQ

Patesco is an independent tool for Wix site owners. ChatGPT, Perplexity and Gemini are trademarks of their respective owners and are referenced here only to describe which assistants are checked. Patesco is not affiliated with, endorsed by, or partnered with any AI provider or with Wix.