How does a collection agency get recommended by ChatGPT and other AI tools?
AI assistants name agencies they can quote. When we asked ChatGPT, with web search on, 10 questions larger creditors ask, it mostly cited agencies’ own pages for one industry, plus lists and directories. So publish one clear page per creditor type you serve, and state your result in one checkable sentence: what you collected per account reached, on what accounts, against what comparison, over what period. A page that says "industry leading results" gives an assistant nothing to repeat.
New to BestPair? See what it is and how it works
What does an AI assistant need from an agency?
AI visibility is whether an assistant names your agency, and cites your pages, when a buyer asks it a question. An assistant builds its answer from pages it can quote. So the question for an agency is not how to rank, but whether its site holds a sentence worth repeating.
For a creditor choosing an agency, the sentence that matters most is the result. Creditors compare agencies on like accounts, and in our view the gap between them sits mostly in what each right-party contact brings in. So the sentence worth publishing states the result per account reached:
| Part of the sentence | Example of what to say |
|---|---|
| The result | Collected per account reached, or per right-party contact |
| The accounts | Kind of debt, stage of placement, age |
| The comparison | Against the previous agency on the same accounts, or a random control group |
| The period | Months from placement, and when the work was done |
| The limit | One book, one period, not a forecast |
The way our own case study is written shows the shape. In a live test at a North American collection agency, a telecom book of 170,000 accounts was split 50/50 at random before the first call. The half matched by BestPair collected $98.76 per account reached against $79.17 on the usual process, 24.7% more. One book in one month, not a forecast for any other floor (the case study). Every part of the table is there, so an assistant can repeat it without guessing. RPC conversion explains why per contact is the measure that separates agencies, and how to show results to a prospective client covers what makes a result believable.
What did ChatGPT cite when we asked?
On 6 November 2026 we asked ChatGPT, with web search on, 10 questions a larger US creditor might ask about collection agencies, and recorded which pages each answer cited:
| Questions asked about | What the answers mostly cited |
|---|---|
| Hospitals and medical accounts | Many agencies’ own healthcare pages, plus a state government page and a medical association page |
| Banks, credit unions and auto lenders | Agencies’ own pages for credit unions or auto finance, plus federal pages (Treasury, CFPB) |
| Utilities | Agencies’ own utility collection pages |
| State and local government | Agencies’ own public sector pages, plus a state comptroller’s page |
| Large business-to-business accounts | A commercial collection association’s member directory, a list site and agencies’ commercial pages |
| First-party collections for lenders | Pages of agencies and outsourcers offering first-party work |
| The largest agencies, and the top ARM firms | List sites, company-data sites and public company filings |
Test run: 6 November 2026, one run per question. One run is a snapshot, and AI answers change from week to week. The pattern held across the questions: where a buyer asked about one creditor type, the assistant quoted pages written for that creditor type. Where the question was about size, it quoted lists.
Why does ChatGPT name different agencies from Google?
Because Google answers a different search. We pulled Google’s top 20 US results for the phrases the same buyers search, such as "accounts receivable management", "healthcare collection agency" and "national collection agency". The pages that ranked were mostly pages built around a city or state, how-to guides, accounts receivable software vendors and commercial collection specialists. Many of the large consumer agencies had almost no share on those phrases.
Only two agencies stood out both in Google’s results and in ChatGPT’s answers. In our view being found on Google and being named by AI need different pages, and doing well on one says little about the other. How larger creditors find an agency has the search volumes behind this.
What should an agency publish?
In our view, six things, all on the agency’s own site:
- One page per creditor type you serve. Hospitals, utilities, government, banks and credit unions, auto lenders, commercial accounts. Each says who you work for, which accounts you take and how you report.
- A result in one checkable sentence. Built from the table above, and repeated on each creditor-type page that it applies to. Collection agency KPIs defines the measures.
- Answers to buyers’ questions. How placements work, how disputes and complaints are handled, what reports the client gets, what security certifications you hold. One question per heading.
- One version of the facts. The same name, locations, licences and creditor types on every page and listing. Assistants can repeat an old page long after it changed, so update or remove pages that no longer hold.
- Listings where AI looks. Association directories and industry lists, each linking to the matching page on your site, and mentions on trade sites.
- Pages a machine can read. Text that loads without scripts, and Organization and FAQ structured data, so an assistant can tell who you are and what each page answers.
No one can buy or guarantee a place in an AI answer. These steps make an agency easier to quote; they do not promise it will be named.
The same result sentence also carries a proposal. How to stand out in a collection RFP covers how tenders ask for it.
Sources
- BestPair research, 6 November 2026: ChatGPT with web search on, 10 enterprise buyer questions; Google’s top 20 US results on enterprise creditor phrases through the DataForSEO SERP API; US search volumes and AI-assistant estimates through the DataForSEO AI Optimization API (Google average of August and September 2026).
Common follow-up questions
Do creditors use ChatGPT to choose a collection agency?
Some use it to learn who is in the market. DataForSEO’s estimate of monthly AI-assistant searches for "accounts receivable management" (551) is close to Google’s 720, and for "collections BPO" it is 125 against 5 on Google. They are estimates, but they put AI answers on the route to a shortlist.
Why does ChatGPT name different agencies from Google?
Because it rewards different pages. In our test, Google’s top results on enterprise phrases were mostly pages built around a city or state, and how-to guides. ChatGPT, answering a buyer’s question, cited pages it could quote directly: an agency’s page for one industry, or a list of agencies by industry.
How can an agency check what AI says about it?
Ask the main assistants, signed out or in a private window, the questions your buyers would ask: the best agencies for your creditor types and states, and your agency by name. Note who is named, which pages are cited and anything out of date. Repeat monthly, since answers change.
What should an agency do when an AI tool says something wrong about it?
Find the page the answer cites, if it shows one, and correct it at the source: your own site first, then the listing or directory it came from. Make sure your own pages state the right facts in plain words, then check the answer again after a few weeks. Assistants update as the pages they read change.
Does BestPair do AI search work for agencies?
No. BestPair sells one product, predictive pairing for collection floors, which changes only which collector works which account.
See it on the main site
Two minutes
Watch the explainer
What your dialer does with the next account, what pairing changes, and what the Floor Assessment measures. With captions.
Watch the videoYour floor
Try the calculator
Three numbers off your dialer and collections reports give an illustrative figure for your floor. It is at best informed guesswork until your own history is measured.
Open the calculatorWhat you get
See the sample report
The report the Floor Assessment delivers, page by page, built on a fictional floor with synthetic data.
See the report
Would it be a bad idea to spend 45 minutes finding out?
If your shop does not qualify, you will hear it on the call, not after you have paid for a report.
Book a 45-minute fit call