Insights
Why does AI recommend my competitors instead of me?
Usually not because they are better. Because more independent sources describe them, describe them consistently, and describe them in the specific terms a customer's question used.
Updated September 22, 2026 · 10 minute read
Short answer
Almost always because there is more corroborated information about them, not because they are better at the work. AI answers are assembled from what the web says about a market, and a competitor who appears in more independent places — described consistently, in the same words customers use — is easier to name with confidence than a business whose evidence is thin, inconsistent, or confined to its own website. The fix is to find out which sources are actually behind the answers, then close the gaps that matter.
It is probably not about quality
This question usually arrives with some heat behind it, and understandably. A business with twenty years of work, better reviews and better outcomes watches an assistant recommend a competitor they consider inferior. The instinct is that the machine is wrong.
The machine is not judging quality. It has no access to your workmanship, your callbacks or how your crews treat a customer's lawn. It has access to what has been written down. When it names three businesses, it is naming the three it can describe most confidently — which correlates with how much independent, consistent, specific information exists about them, and only loosely with how good they are.
That is a frustrating thing to be told and a hopeful one. Quality is hard to change. Evidence is not.
The five reasons we actually find
Across the baselines we run, the reason a competitor appears and a client does not is nearly always one of five things — often several at once.
- They are corroborated in more places. The competitor appears on an association register, a chamber listing, two industry directories and a local news piece. The client appears on their own website. One of those is a body of evidence and the other is a claim.
- They are described consistently. Same name, same address, same phone, same services, everywhere. The client is on record under three name variants with two phone numbers, and a system trying to decide whether that is one business or several may simply not name any of them.
- They use the customer's words. The customer asked about “emergency tree removal after a storm”. The competitor has a page with that phrase and a description of what it involves. The client offers “24/7 storm response solutions” and never says the plain thing.
- They are reachable and the client is not. The client's site blocks AI crawlers, or is not indexed, or sets snippet limits that stop anything being quoted. Nothing downstream matters if the page cannot be fetched or used.
- They have been around the corpus longer. For answers given without a live search, the model is recalling an impression formed during training. A competitor described across the web for a decade has a deeper footprint there than a business that rebranded last year. This is the slowest one to change, and the only one where patience is genuinely part of the answer.
The citations tell you where to look
When an assistant searches before answering, it usually shows its sources. Most people read those citations as references. They are better read as a map of what influenced the answer — and, across enough answers, of which domains hold power in your market.
Run the same set of buyer questions repeatedly and the cited domains stop looking random. A handful appear again and again: a particular directory, an association register, one review platform, occasionally a single local news outlet. Those are the sources that are, in practice, deciding who gets named in your category and your area.
Then the question becomes concrete and answerable: which of those sources feature your competitors and not you?
| Source | Competitors present | You | What it would take |
|---|---|---|---|
| Industry association register | 3 of 4 | Absent | Membership, then a complete profile |
| Regional trade directory | 4 of 4 | Listed, wrong category | Correct the category and services |
| County chamber of commerce | 2 of 4 | Absent | Join, complete the listing |
| Major review platform | 4 of 4 | Present, 9 reviews | Systematic review capture after jobs |
| Local news coverage | 1 of 4 | Absent | Opportunistic; not plannable |
The shape of a source gap analysis
Illustrative, not measured
The table above shows the shape of the analysis, not results from a real client. Greenbug does not publish numbers that did not come from recorded observations — every figure on this site traces back to stored AI responses, and where we have not measured yet, we say so.
Sometimes you are named — and the description is wrong
Worth separating out, because it is the finding that surprises owners most and it is not the same problem as absence.
An assistant names your business and then says something untrue about it. A service area you pulled out of years ago. A service you no longer offer, or one you never did. Hours from a previous owner. A phone number that rings somewhere else. Occasionally a conflation with a similarly named business in another state.
This is worse than being absent. Absence costs you a chance. A confident, wrong description actively misroutes customers, and it will be repeated for as long as the information behind it stands uncorrected. It usually traces back to a stale directory listing or an abandoned profile that nobody has looked at since it was created.
It is also the most fixable thing on this page. Find the wrong source, correct it, and the error stops propagating through the retrieval path fairly quickly.
What not to do about it
The frustration this creates makes a few bad responses tempting, and the market is already selling all of them.
- Buying directory placements in bulk. Low-quality directories are not among the sources that influence answers, for the same reason they stopped influencing search. Fifty listings on sites nobody cites adds nothing.
- Publishing volume for its own sake. Forty thin pages about your town does not make you more recommendable. One page that genuinely answers what a buyer asks is worth more than all of them, because only one of them contains something quotable.
- Trying to prompt-engineer the assistant. Text aimed at manipulating a model — hidden instructions, keyword stuffing for machines — is both ineffective and the kind of thing platforms act against. It is also, straightforwardly, an attempt to deceive the customer reading the answer.
- Paying for a guaranteed recommendation. There is no product that does this. What is being sold is either ordinary work with a guarantee attached that cannot be honoured, or nothing at all.
- Naming competitors on your own pages to ride along. Comparison pages written honestly are legitimate. Pages that exist only to attach your name to a competitor's are transparent to readers and unlikely to produce the association you want.
What closes the gap
Unglamorous, in this order, because the order matters — there is no point earning citations for a business a system cannot resolve.
- Make yourself retrievable. Check crawler access, indexing and snippet directives. This is a day of work and it is the floor for everything else.
- Make yourself one business. Reconcile name, address, phone, service area and service vocabulary everywhere they appear. Correct the stale listings you find on the way; those are often the source of the wrong answers.
- Close the source gaps that are real. Not every gap is worth closing. The ones worth closing are where a source repeatedly appears in the citations for your market, carries genuine authority, and features competitors but not you.
- Publish the things buyers ask about. Pricing ranges with the variables named. What the work involves. When it is urgent. Real evidence with dates and places attached.
- Measure against frozen questions. The same buyer questions before and after, repeated, across platforms, with every response stored. Otherwise you will never know which of the above did anything — and neither will anyone you pay.
Nobody can promise the outcome
None of this guarantees that an assistant will recommend you, and we will not say otherwise. It improves the information these systems draw on and makes it more likely your business is identifiable, accurate and corroborated. Whether the answers change is something we measure and report, including when they do not.
Where to start if this is you
Start by finding out who is actually being recommended and why, rather than assuming. Baselines regularly overturn the assumption: the competitor an owner is worried about turns out not to be named either, and the business appearing instead is one they had not considered a rival.
A baseline records which businesses are named across your buyer questions, how often, on which platforms, and which sources the assistants leaned on. That last part is what turns a complaint into a plan — a specific list of places where your competitors are corroborated and you are not.
Common questions
Does AI recommend competitors because they pay for it?
No. There is no paid placement inside an organic AI recommendation. A competitor appearing instead of you reflects the information available about them, not a purchase.
My competitor has worse reviews. Why are they named and not me?
Star ratings are a small part of it. What tends to matter more is how many independent sources describe the business, how consistently, and whether the text of reviews and listings describes specific services and places an assistant can quote.
How do I find out which sources are influencing the answers?
Run a fixed set of buyer questions repeatedly across platforms and record the cited domains for every answer. Domains that recur are the ones with influence in your category and area. That is what a source gap analysis is.
AI says something incorrect about my business. Can that be fixed?
Usually, and it is often the highest-value fix available. Inaccuracy generally traces to a stale listing or abandoned profile. Correcting the source stops the error propagating through the retrieval path, though answers given from training data take longer to change.
How long before this changes anything?
Corrections and new corroboration can affect answers that involve a live search within weeks. Answers drawn from training data depend on future training runs, and nobody can give you a date for that. We measure both rather than predicting either.
Sources
Primary documentation from the platforms themselves, cited where this page makes a claim about how they behave. Where something is not documented publicly, this page says so rather than asserting it.
- Google Search Central — AI Features and Your Website →
Confirms that indexing and snippet eligibility govern inclusion in Google's AI answers, and that snippet directives limit what can be used.
- OpenAI — Overview of OpenAI Crawlers →
Establishes the separation between search indexing and model training, which underlies the retrieval-versus-recall distinction in this article.
- Perplexity — Perplexity Crawlers →
Documents indexing for search results, the path that makes cited sources observable and therefore analysable.
More resources
- What is AI visibility?
AI visibility is whether AI assistants name your business when someone asks them who to hire. It is a different question from search ranking, it is measured differently, and it can be tested.
- How do I get my business recommended by ChatGPT?
There is no submission form and no way to guarantee it. What you can do is make sure ChatGPT can reach your site, that your business is described consistently everywhere it appears, and that independent sources corroborate it — then measure whether any of it changed the answers.
Find out what AI currently says about your business
A baseline records which businesses are named across your buyer questions, on which platforms, and which sources the assistants relied on. Nothing on this page is a substitute for measuring your own.