Why an AI assistant recommends your competitor instead of you
The recommendation now happens before the search does — and it’s assembled from a public record most owners have never read. Here are the seven scores that decide it, and our own audit, published warts first.
By Greg Bush · Updated July 22, 2026 · 9 min read
An AI assistant recommends whichever business the public record describes most clearly and most consistently — not whichever business is best at the work. The answer is assembled from listings, reviews, your website’s actual text, and third-party pages that discuss your category. If a competitor is described in more places, more consistently, and in language that matches the question, they get named and you don’t. Being better is not an input the model can see.
Something changed in how customers find local businesses, and most owners have not yet seen the consequence. The question “who should I hire for this?” used to produce a page of links to compare. Increasingly it produces a short list of names — three or four — delivered as an answer, with no page two and no obvious way to appeal.
If your name is not on that list, nothing tells you. There is no ranking to watch fall, no traffic dip to investigate, no notification. The customer simply never learns you exist, and you never learn they were looking. That is a genuinely new failure mode, and it is invisible by construction.
How does an AI assistant decide which businesses to name?
Not the way people assume. The assistant is not evaluating quality, and it has no opinion about your craftsmanship. It is assembling a summary from whatever public material describes businesses like yours, and then naming the ones that material describes clearly enough to be nameable.
Four things carry most of the weight:
1 · Whether your basic facts agree with each other
Your name, address, phone, hours, and category appear in dozens of places — your site, Google, Apple, directories, aggregators. When those copies disagree, a system trying to state a fact about you has to choose which version to trust, and the safe move is to say less about you, or nothing. Consistency is not a “best practice” here; it is the precondition for being described at all.
2 · Whether your own pages answer the question being asked
Most small-business websites are written as brochures — a headline, an adjective, a call button. Brochures make poor source material. If someone asks how much a service typically costs, or how long it takes, or what distinguishes two approaches, the assistant needs a passage that actually answers that. If your site never answers it and a competitor’s does, the competitor becomes the source — and sources get named.
3 · What other people’s pages say about your category
A meaningful share of what gets quoted about any local category is not written by the businesses in it. Forum threads, community discussions, roundups, and video transcripts all feed the answer. This is why a business can be excellent and invisible: nobody has written it down anywhere the machine reads.
4 · Whether your reputation looks tended
Review volume and — more than people expect — whether anyone replies. A business with a handful of glowing reviews and no responses reads, to a summarizer, as possibly closed. Sentiment is easy to extract; recency and responsiveness are what suggest a going concern.
Notice that none of the four is about being good at the job. That is the uncomfortable core of this: the assistant is summarizing the public record, and the public record is a different artifact from your business. You can be the best operator in your market and have a thin record, and the assistant will faithfully report the thin record.
The seven scores that decide it
When we audit a business with Footprint, we score seven categories — the same seven for the business and for every competitor, computed by deterministic rubric so the same inputs always produce the same number. These are the levers that actually move whether you get named.
| Category | What it measures | Why an assistant cares |
|---|---|---|
| Technical SEO | Whether your site can be crawled, parsed, and understood at all. | A page that can’t be read cannot be quoted. This is table stakes and usually the easiest to fix. |
| Content | Depth and substance — words per page, thin pages, whether images carry alt text. | Thin pages give a summarizer nothing to lift. A brochure page is functionally invisible as a source. |
| AEO readiness | Question-shaped headings, answer-first passages, and structured data describing what you are. | This is the most direct lever. Content shaped like an answer gets used as one. |
| Local / NAP | Whether name, address, phone and category agree across every source, plus review volume and response rate. | Contradictory facts make you unsafe to state. Unanswered reviews read as inactive. |
| Social presence | Whether linked, active profiles exist at all. | Profiles corroborate that the business is real and current, and add places your name co-occurs with your category. |
| Performance | Load speed and Core Web Vitals, measured on mobile. | Automated fetchers are less patient than human visitors. A slow page is sometimes simply not retrieved. |
| Ads activity | Whether you show up in the public ad-transparency record. | Weakest direct signal of the seven, but it reveals what competitors are actively pushing — useful intelligence even when it doesn’t move your score. |
We ran this on ourselves. We scored 47.6 out of 100.
It would be convenient to illustrate this with an anonymous client who was doing badly. Instead, here is our own audit, run in July 2026 on blackfrog.ai — a firm that sells competitive intelligence for a living.
out of 100
category that cited us
domain ranked for
The detail was worse than the headline. Our content score was 27 out of 100 — an average of 175 words per page, with nine pages under 150 words. Our social presence scored 0, because we had no linked profiles at all. Our mobile largest-contentful-paint was 4.4 seconds. We had four lifetime reviews, rated 4.8, and had replied to none of them.
Meanwhile the technical foundation scored a perfect 100 and we ranked first in the local map pack. That combination is the typical shape of this problem: nothing is broken. The site is fast to build, clean, and says almost nothing that a machine can use. We were a well-built brochure.
We published the number and started working through the list. Alt text and meta descriptions first, then structured data identifying us as a local business, then an answer layer — question-shaped sections on thirteen pages, written to answer the question rather than tease it. The article you are reading is itself part of that work, which seems worth admitting out loud.
What actually moves the needle, in order
Ranked by effect per hour of effort, based on what the rubric rewards and what we found fixing our own:
Fix your facts everywhere (an afternoon)
One canonical version of name, address, phone, hours, and category — then make every listing match it exactly. Free, unglamorous, and the highest-yield thing on this list, because everything downstream depends on your identity being unambiguous.
Reply to every review you have (an hour)
Every one, including the old ones and especially any negative one. Volume matters, but response rate is the signal that a person is home. Then ask each completed customer for a review as a standing habit.
Answer five real questions on your own site (a weekend)
Pick the five questions customers actually ask before hiring you — the ones about cost, timeline, process, and what makes one approach different from another. Give each a heading phrased the way a person would say it, and a direct answer underneath in plain language. Answer the question in the first sentence rather than building to it; a summarizer takes the passage, not the page.
Show up where your category gets discussed (ongoing)
Community forums, local press, trade publications, podcasts. Being mentioned by other people in the context of your category is the signal that is hardest to fake and hardest to buy, which is exactly why it carries weight. Participate honestly and under your own name; the manipulative version is detectable and counterproductive.
Make the site fast on a phone (a day of real work)
Compress and modernize images, preload what renders first, stop loading things below the fold up front. Worth doing for humans regardless; the crawl-reliability benefit is a bonus.
What we would not spend money on: buying links, publishing volume for its own sake, or any product promising to “optimize you for AI” through a mechanism it won’t explain. The levers above are unglamorous and mostly free, which is precisely why they are undersold.
Common questions
How do I check what AI assistants say about my business right now?
Ask them, in a private or logged-out session so your own history doesn’t colour the answer. Use the phrasing a customer would — “who’s a good [your trade] in [your town]” — rather than your business name, since asking about you by name guarantees you get named. Do it across two or three assistants; they draw on different sources and disagree more than you’d expect. Or have us run the check and score it against your competitors.
How long before fixes show up in AI answers?
Longer than search, and less predictably. Listing corrections propagate in days to weeks. New content has to be crawled, then actually selected as a source, which can take a month or more and is not guaranteed. Anyone quoting you a specific timeline is guessing. The honest framing is that these fixes are worth doing because they’re the same things that make you findable and credible to humans, with AI visibility as the compounding second-order benefit.
Is there a way to pay to appear in AI answers?
Not for the organic recommendation itself, as of mid-2026. Assistants are adding advertising surfaces, but those are labelled placements next to the answer rather than a way into it. If someone offers to buy your way into the recommendation, that is either a misunderstanding or a misrepresentation.
Does my industry even come up in AI answers?
Local services, trades, restaurants, and professional services all do — these are exactly the “help me choose” questions assistants are used for. The specifics vary a lot by category and by town, which is the sort of thing worth measuring for your own market rather than assuming either way.
Can't I just add schema markup and be done?
Structured data helps — it’s a meaningful part of the AEO-readiness score — but it describes content that has to exist first. Marking up a page that says nothing tells a machine, precisely and unambiguously, that you have nothing to say. Substance first, markup second.
What to take away
Assistants summarize the public record. They cannot see that you’re better at the work.
Four levers dominate: consistent facts, pages that answer real questions, third-party mentions, and a tended reputation.
Seven categories get scored — technical SEO, content, AEO readiness, local/NAP, social, performance, ads.
We audited ourselves and scored 47.6/100 with a perfect technical foundation. A clean site that says nothing is the standard failure.
The highest-yield fixes are free and boring: correct your listings, reply to your reviews, answer five real questions.
Find out what it says about you.
We’ll point Footprint at your market — your business and the competitors you name, scored on the same seven categories, reviewed by a person before it reaches you. Free, once, no pitch attached.
Practical AI for small and mid-sized business. Live products, verified numbers, and your experts in charge of every decision.