Néstor Romero answers: why AI recommends one Panamanian business and not another

Néstor Romero answers: why AI recommends one Panamanian business and not another

In short: when an artificial intelligence assistant recommends a business, it does so in three steps (it searches, it chooses, it writes) and the name that appears comes mostly from what third parties say about that business, not from what the business says about itself. So explains Néstor Romero, professor of artificial intelligence and digital marketing at Quality Leadership University, in a telephone conversation with Radar Panamá. His measurements: in one entire sector of a Latin American country, 46 out of 46 commercial searches returned an AI answer before the links; in the United States, clicks fall by almost half when that summary appears; and across more than 3,000 references cited by assistants, social networks contributed not a single citation. For Panama the figure does not exist, and anyone who offers it “is making it up”. His 90-day recipe: check the robots.txt, one page per service, and reviews with text.
Contents of this interview
  1. Who answers and why it matters
  2. What happens behind the scenes when AI recommends a clinic
  3. The figure that does not exist for Panama
  4. Simple website and active Instagram: the bad news
  5. What others say weighs more
  6. A Spanish-only website: it depends on who the customer is
  7. The costliest mistake: solving the wrong problem
  8. Google reviews: they read them, but not all systems alike
  9. Three things in ninety days
  10. What a Panamanian SME can do this week
  11. Frequently asked questions
  12. Sources consulted

Who answers and why it matters

Néstor Romero is an engineer from the University of Córdoba (Spain) and holds a doctorate from the University of Málaga. He is president of the European Digital Marketing Association and vice-president of the European Neuromarketing Association. He teaches artificial intelligence and digital marketing at Quality Leadership University (Panama) and has been a guest lecturer at the University of Louisville, the University of Málaga, the University of Chile and UNED, with more than 20 years of digital projects in more than 30 countries. On 23 September he spoke at the LiderIA Summit, in the auditorium of the Torre de las Américas in Panama City, with the lecture “Communicating in the age of AI: brands, audiences and reputation in the new digital economy”.

Deilys Romero, economy and business writer at Radar Panamá, spoke with him by telephone on 26 September, days after his talk at the LiderIA Summit, with the professor in Panama City. The sentences in quotation marks are verbatim from that conversation. He set one caveat at the outset: where a figure does not exist, he says so rather than estimating it.

What happens behind the scenes when AI recommends a clinic

“Three things happen in a row, and you can lose out at any one of them. First the machine searches: it retrieves a dozen pages that look relevant. Then it chooses: it keeps three or four, and at this point it no longer asks whether they deal with the subject, but whether they strike it as reliable and clear. And finally it writes, in its own words, deciding whom to name.”
Néstor Romero, professor at Quality Leadership University

1. It searches

It retrieves around a dozen pages that look relevant.

2. It chooses

It keeps three or four. It no longer asks whether they deal with the subject, but whether they are reliable and clear.

3. It writes

It writes in its own words and decides whom to name. A business can drop out at any of the three phases.

Romero explains that the name that appears “usually comes from what third parties say about that clinic (reviews, directories, articles), not from what the clinic says about itself”. And he adds a prerequisite that almost nobody meets: the website must state, in readable text, what it does, where and at what cost.

“If that is not written down, it is not that the machine does not choose it: it is that it cannot evaluate it.”
Néstor Romero

The figure that does not exist for Panama

Radar Panamá asked him what percentage of commercial searches in Panama are already resolved in an AI answer. The reply was, in his words, uncomfortable.

“That figure does not exist for Panama, and whoever gives it to you is making it up. Nobody publishes how many times something is asked of an assistant, and there is no national measurement of commercial searches resolved without a click.”
Néstor Romero
Key figure: in one entire sector of a Latin American country, 46 real commercial searches returned an answer generated by artificial intelligence before the links: 46 out of 46. In the United States, where real behaviour is measured, clicks fall by almost half when that summary appears. Néstor Romero’s own measurement.

According to Romero, the phenomenon “moves faster where the decision is comparative: professional services, health, education, banking and hospitality”. Those are the sectors Radar Panamá has put to the test in its Radar AI series, with studies on which dental clinics AI recommends in Panama and which immigration lawyers AI recommends.

Simple website and active Instagram: the bad news

Does an SME with a simple website and an active Instagram account stand a chance? Romero says yes, “every chance”, but first puts forward a finding from his own research.

“In our own study of more than three thousand references cited by assistants, the four big social networks contributed not a single citation. Not one. It is not contempt: it is that they block automated crawling. That Instagram page, for artificial intelligence, does not exist.”
Néstor Romero

From there, he proposes a specific order.

  1. Check that the website can be read. That it is not blocked and that the text really exists, not inside images.
  2. Publish on the business’s own domain what is currently somewhere else. Prices, services, opening hours, who it is for and who it is not for.
  3. The reviews. In his words, “they are the only thing a third party writes on your behalf”.

What others say weighs more

“What others say weighs more, and for a sensible reason: nobody speaks ill of their own business, and the machine knows it.”
Néstor Romero

The example that best illustrates it, he says, he saw in industrial air conditioning: the market leader had a worse website than much smaller rivals and, even so, artificial intelligence recognised it without hesitation as first in the sector. The reason lay outside its website: dozens of locations with business listings full of real reviews.

“And the striking thing: some of those listings it had not even claimed. They were orphan listings, built by its own customers. It did not have a good story about itself. It had something better: a multitude of third parties attesting that it existed and worked.”
Néstor Romero
What counts for an AI to cite a businessWhat Romero saysWho controls it
Readable text on the business’s own websitePrerequisite: what it does, where and at what cost, in text rather than imagesThe business
Reviews on business listingsWhat weighs most; the only thing a third party writes on the business’s behalfCustomers
Directories and articlesWhere the name the machine cites usually comes fromThird parties
Social networksZero citations across more than 3,000 references; they block crawlingDoes not count for AI

A Spanish-only website: it depends on who the customer is

“It depends entirely on who you sell to. If your customer is Panamanian, Spanish is what matters and English will not save you anything. If you sell to foreign investors, patients or buyers, then it is not a luxury: it is half your market.”
Néstor Romero

And he offers a fact that, he says, almost nobody knows. In a systematic review he published with several colleagues, of thirty studies on this subject not one evaluates whether what works in English works the same in another language.

“Everything the industry claims has been measured in English. Anyone operating in Spanish is applying conclusions that nobody has verified in their language.”
Néstor Romero

The costliest mistake: solving the wrong problem

“The costliest is not spending badly: it is solving the wrong problem.”
Néstor Romero

Romero recounts the case of a company that called him convinced that its drop in traffic was due to artificial intelligence: “the fashionable explanation, and it fitted everything”. On reviewing nine months of data, impressions had fallen by around 40% and clicks by 75%. That disproportion was the diagnosis: they had not stopped appearing, they had stopped appearing at the top.

“There was no machine involved. They had spent months preparing for the wrong problem while bleeding out on the usual one. Before investing in the new, check that you are not bleeding out on the old.”
Néstor Romero

Google reviews: they read them, but not all systems alike

“They read them, but not all systems do, nor in the same way. In our own measurement of businesses in a single neighbourhood, one assistant placed two establishments relying on business listings and their reviews (one with more than nine hundred and a near-perfect rating); another placed none, because it was reading editorial lists instead of listings. Nine hundred reviews won in one place and bought nothing in the other.”
Néstor Romero

On whether the text matters or only the rating, Romero holds that a detailed review hands the machine something it can use: what was done, how much it cost, how it went. “Ten stars without words only provide an average,” he sums up, with a caveat he asked to keep: “I have no measurement of my own that quantifies it, so take it as professional judgement and not as data.”

Three things in ninety days

The first, according to Romero, comes before anything else.

“Type into your browser your domain followed by a slash and the words ‘robots dot txt’. If you see a line that forbids entry to all search engines, drop everything and call whoever runs your website. I found it at a clinic on the verge of closing, and it was fixed in a minute.”
Néstor Romero

The second is to create one page for every service sold, with the price or range, the timeframe, the coverage and who it is not the right option for. “That last part looks commercially suicidal and it is exactly what allows the machine to recommend you to those who do fit,” he explains. The third: claim the business listings and ask ten satisfied customers for reviews with text.

The three 90-day actionsWhat it involvesWhy it matters
1. Check the robots.txtOpen domain/robots.txt and check that no line forbids entry to all search enginesIf it is blocked, nothing else works; it is fixed in a minute
2. One page per servicePrice or range, timeframe, coverage and who it is not the right option forSaying who it is not for lets the machine recommend the business to those who do fit
3. Listings and reviews with textClaim the business listings and ask ten satisfied customers for written reviewsIt is the only thing a third party writes on the business’s behalf, and what weighs most

What a Panamanian SME can do this week

The conversation yields tasks with no budget that, according to Romero, decide whether a business can even be evaluated by an assistant. First, check the domain’s robots.txt file; Google’s official documentation explains what each line means. Second, check that services, opening hours and coverage areas are written on the website as text, not only in images or on Instagram. Third, locate and claim the Google Business Profile listing, including any that customers themselves may have created. And no new investment before confirming, with data, that the drop in traffic does not come “from the old”. For those registering a new company, Radar Panamá’s guide to opening a company in Panama covers the preliminary steps.

More Radar Panamá interviews: Néstor Romero answers: which jobs AI changes first in Panama and what universities are missing · Kathia Rivera answers: incorporating a company in Panama as a foreigner, without travelling and in a week

Frequently asked questions

How does an AI decide which business to recommend?

According to Néstor Romero, in three steps: it searches a dozen relevant pages, it chooses three or four depending on whether they strike it as reliable and clear, and it writes, deciding whom to name. That name usually comes from what third parties say about the business (reviews, directories, articles), not from what the business says about itself.

What percentage of searches in Panama is already resolved by AI?

That figure does not exist for Panama, and Romero warns that anyone offering it is making it up. What he has measured: in one entire sector of a Latin American country, 46 out of 46 commercial searches returned an AI answer before the links, and in the United States clicks fall by almost half.

Does Instagram help AI recommend a business?

No. In Romero’s study of more than 3,000 references cited by assistants, the four big social networks contributed not a single citation, because they block automated crawling. For artificial intelligence, an Instagram page does not exist; whatever is there must be published on the business’s own domain.

Which weighs more, the business’s own website or what third parties say?

What others say weighs more, because nobody speaks ill of their own business and the machine knows it. Romero cites an industrial air-conditioning leader with a worse website than its rivals, which AI recognised as first in the sector thanks to its dozens of listings with real reviews, some unclaimed.

Do Google reviews influence what AI recommends?

Yes, but not in all systems nor in the same way. In Romero’s measurement of businesses in a single neighbourhood, one assistant relied on listings and their reviews and another read editorial lists. Reviews with text contribute more than wordless stars, although Romero offers that as professional judgement, not data.

What three things should an SME do in ninety days?

First, check the domain’s robots.txt and, if it blocks all search engines, call whoever runs the website. Second, create one page per service with price or range, timeframe, coverage and who it is not the right option for. Third, claim the business listings and ask ten satisfied customers for reviews with text.

Sources consulted

Who does artificial intelligence already recommend in Panama?
Radar Panamá put AI assistants to the test with real questions from home buyers and measured which names appear and why.

Read the Radar AI study on developers and residential projects

Radar answers

Ask about this article

Type your question and we will show you the answer inside this article or in other Radar Panamá publications.

Was this article useful?
By Deilys Romero

Deilys Romero is a writer at Radar Panamá. A business administrator with a specialisation in accounting, Venezuelan, 48, she covers the economy, companies, taxes and SME management with a practical approach. Based between Panama and Colombia, she closely follows the business activity of both countries.