First, what the machine is actually doing when it answers
When a homeowner types "who's the best electrician in Fort Myers" into ChatGPT or Gemini, the model is not reading a ranked list of ten links and picking the top one. It runs a retrieval step: it pulls a handful of source documents that seem relevant, reads them, and writes a summary that names the businesses those documents support. Perplexity and Google's AI Overviews do this in the open, with citation footnotes you can click. ChatGPT does it more quietly, but the mechanic is the same.
That changes the whole game. A blue-link ranking is a contest between pages. An AI answer is a contest between facts the model is confident about. The model would rather name a business it can corroborate from three sources than a business that ranks #1 but only appears on its own website. Confidence beats position.
Two consequences follow, and both are good news for an established shop. First, being on page one of Google is neither necessary nor sufficient: plenty of #1-ranked contractors never get named because the AI cannot verify anything about them beyond their homepage. Second, the model has no loyalty to whoever spends the most on ads. It names what the evidence supports.
It also matters that the model is risk-averse by design. When a homeowner asks who to call for a burst pipe or a dead panel, the AI is putting its own credibility on the line with a name. If it is not confident, its safest move is to hedge: it names the two or three shops it is most sure about and leaves everyone else out entirely. There is no consolation position four in an AI answer the way there is on a search results page. You are either named or you are invisible, and the line between the two is confidence, not effort.
So the real question is not "how do I rank in AI." It is "what does the model need to see before it feels safe saying my company's name out loud to a homeowner who is about to spend real money." The rest of this guide is that checklist: entity, citation, corroboration.
Entities: can the model tell exactly who you are?
An entity is a thing the model recognizes as a single, specific object in the world: your business, not a category, not a competitor with a similar name. Before an AI can recommend you, it has to resolve you cleanly. That means it can answer, with confidence, what your company is called, what trade you're in, where you work, and what you actually do. If any of those are fuzzy, the model treats you as noise and reaches for a shop it understands better.
Entity confusion is more common than owners think. Two plumbers named "A-1 Plumbing" in the same metro. A roofer whose site says "roofing and restoration and general contracting and remodeling" so the model cannot decide what he is. A shop whose name is written five different ways across the web (with LLC, without, with the city, without). Every inconsistency forces the model to guess, and guessing makes it cautious.
You make yourself resolvable by being boringly consistent and specific. Same legal name, same phrasing, same service list, everywhere. A clear one-line description of the trade and the service area, repeated in the same words on your site, your profile listings, and your structured data. Structured data (schema) is where this pays off hardest: a LocalBusiness or trade-specific type with your name, category, area served, and same-as links to your profiles hands the model the entity on a plate instead of making it infer one.
Specificity is its own advantage here. A shop that says "we do everything" is harder to resolve than a shop that says "we install and service standby generators, period." The model recommends against clear categories, so the plumber who is unmistakably the trenchless sewer specialist gets named for that query more readily than the do-it-all handyman who technically also does it. You are not narrowing your business by naming your lane clearly. You are making yourself the obvious answer to a specific question, which is exactly the question a homeowner is asking the AI.
Here is the test. Ask ChatGPT "what does [your exact company name] in [your city] do?" If it answers correctly and specifically, you're a resolved entity. If it says "I don't have specific information" or describes the wrong company, that is your first repair, and it comes before any citation or corroboration work will move the needle.
Citations: are your own pages clean enough to quote?
Once the model knows who you are, it needs something it can quote. AI answer engines pull short, factual passages out of pages and paraphrase them. A page that states facts plainly gets used. A page that hides its facts inside marketing prose, sliders, or a wall of "we're the best in town" adjectives gives the model nothing liftable, so it moves on.
Citation-worthy pages share a shape. The answer is near the top, not buried. Facts are stated as facts: what you do, where, since when, what it costs to start, who it's for and who it isn't. Questions get direct headings and direct answers. There is an at-a-glance block a machine can parse in one pass. The page loads fast (under 2 seconds) and reads cleanly without JavaScript, because retrieval bots do not wait around.
A few page-level habits that make a shop quotable:
- Answer the question in the first two sentences of the section, then support it. The model lifts the top.
- Use plain declarative facts: "We service standby generators in Collier and Lee County. Since 2011." That sentence can be quoted verbatim. "Your trusted generator experts" cannot.
- Give every real question its own heading and answer it in two or three sentences. That is the exact shape an answer engine wants.
- Put the hard facts in a small table or list: service area, trades, license, hours, response time. Structured beats prose for a parser.
Notice this is a content and structure discipline, not a keyword game. You are not stuffing a phrase. You are writing pages a machine can read a fact off of without getting confused. Do that across your service pages and you become the source the AI reaches for when it needs to say something concrete about a shop like yours.
Corroboration: does the outside world agree with you?
This is the piece owners underestimate, and it is often the deciding one. An AI is far more willing to name a business when the same facts show up in places it doesn't control: directories, review platforms, association listings, local press, supplier and manufacturer pages, permit and licensing records. When five independent sources say the same shop is a licensed roofer in Naples, the model treats that as established fact. When only your own site says it, the model treats it as a claim.
Think of it as the difference between telling the model something and letting the model discover you already told the truth. Your website is testimony. Corroboration is the witnesses. AI answer engines lean on corroboration heavily precisely because it is hard to fake at scale, which is exactly why it earns trust.
What corroboration looks like for a contractor, in rough order of weight:
| Source type | What it corroborates | Why the model trusts it |
|---|---|---|
| Review platforms and profiles | Name, trade, service area, that you're real and active | Independent, high-traffic, hard to fabricate at volume |
| Trade associations and licensing bodies | Credentials, specialty, legitimacy | Authoritative, rarely wrong |
| Manufacturer and supplier "find a pro" pages | What brands and systems you actually install | Vendor has no reason to list you falsely |
| Local news and community mentions | That you operate where you say you do | Third-party, geographically specific |
| Consistent directory listings | Name, address, phone, category alignment | Volume of agreement reduces model uncertainty |
The failure mode is contradiction. If your site says one service area and your listings say another, or your name is spelled three ways, the model sees disagreement and downgrades its confidence in all of it. Corroboration is not just presence in more places. It is the same story told the same way in more places.
One more thing owners get wrong: they chase volume when they should chase alignment. Twenty half-finished directory listings with mismatched names do more harm than four accurate ones that agree with each other and with your site. Before you go add yourself anywhere new, fix the sources that already exist and make them say the same thing. The model is doing a kind of cross-examination, and a witness who contradicts himself hurts your case more than one who never showed up.
How entity, citation, and corroboration stack up together
None of the three works alone. They compound. A clean entity with no citable pages gives the model a name but nothing to say about you. Great pages with a fuzzy entity get quoted for someone else. Strong corroboration for a business the model can't resolve just spreads the confusion wider. The shops that get recommended have all three pointing at the same, specific truth.
Here is the way to read your own gaps:
- The model doesn't know you exist. That's an entity problem. It cannot resolve your name to a real business in a real place. Fix consistency and structured data first.
- It knows you exist but says nothing useful. That's a citation problem. Your pages don't state liftable facts. Rewrite service pages to answer questions plainly and add an at-a-glance block.
- It describes you but won't recommend you. That's usually a corroboration problem. The facts live only on your site. Build agreement across profiles, associations, and vendor listings.
The order matters. Corroboration work is wasted effort if the entity is still ambiguous, because you're reinforcing a blur. Citation work is wasted if the pages are fast and clean but the model can't tell which business the page is about. So the sequence is almost always entity, then citation, then corroboration, and then you measure whether the mention actually appears.
That last step is the one most shops skip. You have to check the answers. Ask the real questions a homeowner would ask, across ChatGPT, Gemini, Perplexity, and Google's AI Overviews, and see whether you're named. The answers move month to month as the models retrain and re-retrieve, so this is a tracking job, not a one-time audit.
Reading the gap is also how you avoid wasting money. An owner who assumes he has a corroboration problem and buys a pile of listings, when his real issue is that the model can't even resolve his name, has spent a quarter reinforcing a blur. Diagnosis before treatment. The three-question read above tells you which of the three layers is actually failing, and every dollar after that goes to the layer that will move a mention.
What a contractor can actually do about it this quarter
You don't need to boil the ocean. The work is concrete and it sits inside this lane: GEO and AEO, the visibility layer that decides AI mentions. It is not your Google ranking and it is not your map pack. Those matter, they feed into this, and they live in their own silos. This is the citation layer specifically.
A realistic first pass, in order:
- Resolve the entity. Lock one exact business name, one clear trade line, one service area, and write them the same way everywhere. Add or fix structured data so a machine can read the entity without guessing.
- Make three or four pages quotable. Your top service pages. Answer up top, facts as facts, one at-a-glance block each, fast and clean without scripts.
- Build corroboration for the same story. Get the same name, trade, and area confirmed across the platforms and listings the model already reads. Fix contradictions before adding new listings.
- Track the answers. Run the homeowner's actual questions across the four engines and log whether you're named. Re-run monthly.
Expect a timeline, not a switch. Entity and page fixes can show up in AI answers within weeks once the engines re-retrieve. Corroboration and competitive terms take longer, generally 4-9 months for anything a lot of shops are fighting over, because you're building agreement across sources you don't control. That is the honest range. Anyone promising you an instant ChatGPT mention is selling something we won't.
The upside for an established shop is that this channel is still quiet. Most contractors, and most of the agencies serving them, are still optimizing for blue links and have no plan for the citation layer at all. The shops that get their entity clean and their pages quotable now are the ones the AI keeps naming while everyone else wonders why ChatGPT still doesn't mention them.