marumba insightsARTICLES & ANSWERS

SEO, GEO & AEO

Grow & be found, GET RECOMMENDED BY AI THE TOP PICK

How AI assistants decide which businesses to recommend.

How AI Assistants Decide Which Businesses to Recommend

By Todd Sullivan · November 9, 2026 · 8 min read

The short answer: an AI assistant recommends a business when it can find reliable, consistent information about it, from more than one place, that matches the question being asked. There is no secret score. The assistant combines what it learned in training, what it retrieves live from a search index, and what it can corroborate across the web. Your job is to make each of those three things easy.

How a generative answer gets built

Most assistants that can answer "who is a good X near me?" do three things. They interpret the question, often breaking it into several related sub-questions (Google calls this query fan-out). They retrieve supporting pages from a search index. Then they write an answer, citing or naming sources that looked relevant and trustworthy. Google's guide for its AI features describes exactly this retrieval-and-fan-out pattern, and says the features draw from the same Search index and quality systems as ordinary results.

The practical consequence: you cannot optimize for the language model alone. You have to be findable by the retrieval step, clear enough for the model to summarize correctly, and corroborated enough to be worth naming.

A worked example: one question, many searches

Suppose someone asks an assistant, "Who is a good electrician in Mesa for an electrical panel upgrade?" Behind that single question, the system may quietly run several narrower ones: electricians serving Mesa, panel upgrade cost, licensed and insured contractors, recent reviews of panel work. It then looks for sources that answer those pieces and writes one response. This example is illustrative, but it shows why a business can appear for one phrasing and vanish for another. You win by being a good answer to several related questions, not one keyword.

It also shows what a business can influence. A dedicated panel-upgrade page that states typical scope, permits and timing helps the cost question. A complete, accurate business profile helps the "serving Mesa" question. Reviews that mention panel work help the reputation question. Each one is a separate chance to be included.

The signals that plausibly matter

1. You can be crawled and indexed

If your pages are blocked, slow to respond or hidden behind scripts, they cannot be retrieved. Check robots.txt, your firewall or CDN bot rules, and whether your key content is present in the raw HTML rather than injected by JavaScript after load. Many crawlers do not run scripts the way a browser does. Assistants differ in which index they draw on (some lean on Google's, some on Bing's or other partners'), so submit your site to both Google Search Console and Bing Webmaster Tools.

2. Your identity is unambiguous

Use one business name, one address format, one phone number and one short description everywhere: your website, Google Business Profile, Apple and Bing maps listings, industry directories and social profiles. Consistency lets a system confidently merge scattered mentions into a single entity. Add Organization or LocalBusiness structured data so your own site states the same facts.

3. Others vouch for you

Reviews, directory listings, local press, association memberships, partner pages and genuine forum discussions all act as corroboration. One SEO firm argues that for AI visibility, your footprint on third-party sites often counts for more than anything on your own pages (iPullRank). That is an opinion rather than a proven rule, but it fits how retrieval works: a claim confirmed in several places is safer to repeat.

4. Your content is specific and original

Google's guidance repeatedly emphasizes unique, non-commodity content. For a service business, that means pricing context, process descriptions, service areas, comparisons, photos of real work and answers to the questions your customers actually ask, not paraphrases of what every competitor says.

5. Answers are easy to lift

Pages that state the answer in the first lines, use headings that mirror real questions and include concrete facts are simply easier for a system to quote accurately. You do not need to chop content into tiny fragments for bots; write clearly for people.

6. Information is current

Show dates, update key pages when facts change, and keep hours, services and pricing details accurate. Stale or contradictory facts make an assistant less likely to trust you.

Local businesses: rankings are not recommendations

If you serve a local market, the evidence so far points to a few consistent themes, with the caveat that most of it comes from marketing firms studying their own samples rather than from the assistant makers.

  • Appearing in the map results does not guarantee an AI mention. In SOCi's 2026 index, brands showed up in Google's local three-pack for about 36% of locations, but ChatGPT recommended about 1%, Gemini about 11% and Perplexity about 7% (Search Engine Land).
  • Assistants lean on different data. The same report found that Gemini, which is grounded in Google Maps, repeated business details correctly far more often than ChatGPT and Perplexity did. Wrong hours, phone numbers and addresses on third-party sites are a real risk.
  • Reviews correlate strongly with visibility. One study of local businesses found those surfaced by both ChatGPT and Perplexity averaged about 133 Google reviews, versus about 11 for businesses neither tool surfaced (Insites). Correlation is not proof that reviews cause mentions, since busy businesses tend to have more of everything, but it supports a steady review habit.

The practical reading: the Google Business Profile, accurate listings and genuine reviews are the cheapest, most broadly useful work you can do for local AI visibility, and they help ordinary local search at the same time.

Common mistakes

  • Assuming a strong Google ranking covers you. It helps, but test the assistants directly.
  • Letting facts drift. A new phone number or moved office updated on your site but not on a dozen directories creates conflicting evidence.
  • Thin service pages. One generic page listing every service gives a system little to quote. Give important services their own specific pages.
  • Buying or faking reviews. Platforms are tightening enforcement, and fake reviews damage trust with real customers and with the systems that read them.
  • Hiding key facts in images or PDFs. Put pricing context, service areas and credentials in plain page text.

What probably does not move the needle on its own

  • Schema markup as a citation trick. Google says no special markup is required for its AI features, and roundups of controlled tests published so far have found no citation lift from schema alone (ThriveStack). Markup is still worth having for clarity and rich results.
  • An llms.txt file. Google says it does not use it, and server-log studies show almost no AI crawler requests for it (more in our llms.txt article).
  • Blocking AI crawlers to "protect" content. It is a legitimate business decision, but be aware of the trade-off. A 2026 study of news publishers found that blocking AI crawlers was associated with a roughly 7% traffic decline over six weeks (PPC Land summary). Your situation may differ.

Be deliberate about crawlers

Assistant companies typically run separate bots for different purposes: one to fetch pages live when a user asks a question, one to build search indexes, and one to collect training data. Their names and behaviors change, so check each vendor's current documentation before editing robots.txt. A reasonable default for most businesses that want to be recommended is to allow the search and retrieval bots, and decide separately about training bots.

Test it yourself: the 10-question audit

  1. Write ten questions a real customer would ask, such as "best [service] in [city]," "how much does [service] cost," and "who can help with [problem]."
  2. Ask each question in two or three assistants, in a fresh chat each time.
  3. Log whether your business is named, whether the facts are right, and which sources are cited.
  4. Repeat on a different day. Answers vary from run to run, so a single test proves little.
  5. Look at who does appear. Their profiles, reviews and pages show you what the assistant considers trustworthy.
  6. Fix the cheapest gaps first: wrong facts, missing listings, thin pages.

Your first 30 days

  1. Days 1 to 3: run the 10-question audit and save the results in a spreadsheet.
  2. Days 3 to 7: claim and complete your Google Business Profile and Bing Places listing. Check categories, hours, service areas and description, and add current photos.
  3. Days 7 to 14: audit your name, address and phone across the top directories and fix mismatches.
  4. Days 14 to 21: check robots.txt and firewall rules for accidental bot blocks, and confirm key pages load without JavaScript.
  5. Days 21 to 30: upgrade your two most important service pages with direct answers, specifics and proof, and start a simple routine of asking each satisfied customer for a review.

Then re-run the audit and compare. Even a small improvement in accuracy counts as progress.

Honest limits

No one outside these companies can tell you exactly how recommendations are chosen, and the systems change often. Anyone guaranteeing placement is guessing. What you can do is remove obstacles, publish things worth citing, and build a consistent, corroborated presence. Those steps help ordinary search, too, so the work rarely goes to waste.

Quick answers

Can I pay to be recommended?

Organic recommendations are not for sale, and no one can sell you a guaranteed mention. Advertising formats inside AI products are evolving and differ by platform, so check each vendor's current documentation before spending.

How often should I re-test?

Monthly is plenty for most small businesses. Answers fluctuate, so compare trends across several runs instead of reacting to one result.

Do my reviews need to be on Google?

Google reviews appear to carry the most weight, especially for assistants tied to Google's data, but reviews on Yelp, industry platforms and other trusted sites also add corroboration. Prioritize the platform your customers actually use.

What if an assistant says something wrong about my business?

Find where the wrong fact originates, usually an old listing, directory or page, and correct it at the source. Updating your own site helps, but the stale copy elsewhere is often what is being repeated.

Sources

Want help with this?

Tell us what you are working on and we will point you in the right direction.

Call Get a quote