How AI assistants decide which companies to recommend

The three-step decision

Strip away the interface and every assistant answers a "recommend me a vendor" question the same way. I have spent two years watching this machinery from both ends, as a consultant fixing companies' visibility and as someone the assistants describe, and the three steps below hold across every engine I test.

Step one: retrieve. The assistant turns the question into searches, often several at once, and pulls back a shortlist of pages: listicles, directories, vendor sites, forum threads. If nothing about you is retrieved, you are out before anything intelligent has happened. Retrieval runs on search indexes, which is why classic SEO plumbing still gates everything, and why a blocked crawler is fatal (see the robots.txt guide).

Step two: cross-check. The model weighs what it retrieved. Do multiple independent sources describe this company consistently? Does the vendor's claim about itself match what others say? Is the information current? Companies that appear once, describe themselves one way while the internet says another, or last got written about in 2019, get discounted here.

Step three: compose. The assistant writes its answer from the surviving material, quoting or paraphrasing the passages that were easiest to lift. If your site states plainly what you do, for whom, and where, your own words often become the description. If it does not, the machine composes your description from whatever else it found, and you live with the result.

Query fan-out: the buyer asks once, the machine asks a dozen times

The step-one detail that changes strategy most. Modern engines, Google's AI Mode most explicitly, decompose one question into many parallel sub-queries. "Suggest an ERP implementation partner for a mid-size Pune manufacturer" becomes separate searches about ERP partners in India, implementation costs, manufacturer case studies, comparison threads, and more, each retrieving its own results before the answer is composed.

The consequence: a company with one good page gets one lottery ticket, while a company whose site and footprint cover the surrounding questions (costs, process, comparisons, industries served) holds tickets in several draws for the same buyer question. This is why topical completeness beats a single monolithic page, and why thin "we are the best ERP partner" brochure sites, whatever their design budget, hardly ever surface: they answer none of the sub-questions the machine actually asked.

Where each assistant looks

The mechanics above are shared; the retrieval sources are not. Citation overlap between engines is small, roughly one answer in ten, so it pays to know who reads what.

Two practical consequences. First, submit your sitemap to Bing Webmaster Tools; it takes ten minutes, it feeds ChatGPT and Copilot, and in India almost nobody does it. Second, use Perplexity as your test bench: publish, wait a week, ask again, and you have real evidence of whether your content is citable.

The cross-checking step is where Indian SMEs lose

Most founders I speak to assume the fix is on their website. Usually the website is the healthiest part of their footprint. What is missing is the agreement layer around it: the category listicles that rank for "top X in India" and never mention them, the directory profiles with a defunct address, the review platforms with three reviews from 2020, the total silence on Reddit and Quora where their buyers ask questions.

Assistants read that layer as the jury. A company whose website says one thing while the surrounding record is thin, stale or contradictory does not get recommended; it gets hedged, or skipped. I know this failure mode personally: my own company spent years being described by assistants with positioning we had abandoned, because the off-site record still carried it. The dull work of getting listed, described consistently and reviewed recently moves recommendations further than another website rebuild ever will.

The recommendation-readiness checklist

Score yourself one point per yes. In my experience companies below six are effectively invisible to category questions, whatever their Google rankings.

1. Your home page answers "what do you do, for whom, where" in its first screen of text.
2. The same one-line description of your company appears on your site, LinkedIn, and every directory profile.
3. Your structured data declares your Organization with a stable identity, and it matches the prose (see structured data).
4. You appear in at least two third-party pages that rank for your category's "best X" or "top X" searches.
5. You hold reviews on at least one platform an assistant can read, with at least one from the last six months.
6. Something dated within the last year exists about you: a news mention, a published article, an updated page.
7. Your site covers the questions around your service (cost, process, comparisons), not only the service itself.
8. Your sitemap is submitted to both Google Search Console and Bing Webmaster Tools.
9. All retrieval crawlers are allowed at every layer: robots.txt, CDN, plugins.
10. Nothing machine-readable contradicts anything else: no old address in one directory, abandoned positioning in another.

What moves the needle, and how fast

Timelines vary with crawl frequency and competition, but these are the lags I plan around with clients:

What does not work

Because the field is young, it is full of confident nonsense, and some of it is sold at retainer prices. Things I have tested or watched fail: stuffing FAQ schema onto pages in the hope of being "chosen"; adding llms.txt and calling the job done (Google says outright that Search ignores it); hidden text addressed to AI models, which is retrieval-invisible at best and trust-damaging at worst; bursts of fake or incentivised reviews, which the cross-checking is increasingly built to discount; and swapping the publish date on unchanged content, which Perplexity in particular sees through as soon as it fetches the page. None of these create independent agreement about you, and independent agreement is the only thing being measured. I wish there were a shortcut to sell you. There is not, and the people who claim otherwise know it too.

What to do this quarter

Baseline first: run the query panel exercise and write down what each assistant says about you and your category today. Then work the checklist above from top to bottom, cheapest first: your own site's one-line clarity, profile consistency, Bing submission, then one or two category listicle inclusions and review recency. Recheck monthly. Expect Perplexity to move first and ChatGPT's view of you to move last, for the reasons in the tables above. The mechanics are stable even while the interfaces churn, and the compounding is real: every consistent, current, independent trace of your company makes the next recommendation decision easier to win.