And why this completely changes the rules of GEO promotion
We spent several hours talking with Perplexity AI, deliberately drawing out its decision-making logic: why it recommends some companies and ignores others, even when they are larger and better known. The result exceeded expectations. It turned out that the algorithm is already working in a more sophisticated way than most SEO specialists assume — while also having quite specific vulnerabilities that can be exploited.
Below is the complete mechanics of how Perplexity (and similar AI search engines) decide whether to make a recommendation. And what exactly needs to be done to get on that list.
Factor 1. A narrow focus matters more than broad coverage
The first thing Perplexity said outright: it prefers specialized companies. Not those for whom certification is one of a hundred items on the price list, but those who have built an entire website around that service: sections, FAQs, case studies, articles, and calculators.
The logic is simple: the neural network evaluates the depth of expertise through the volume of content on the topic. The more pages, sections, and materials devoted to a specific service, the more confident the model is that the company truly specializes in it.
Practical takeaway: if you want to appear in AI recommendations for a specific service, you need more than just a landing page—you need a developed thematic cluster of pages. This is how the websites most frequently recommended by Perplexity are structured.
Factor 2. External proof matters more than the website itself
The second criterion turned out to be even more important than the first. Perplexity said directly: one website is not enough. It needs external sources—other websites where the brand is mentioned and specifically the services related to the query are described.
And here is the key revelation from the conversation:
“What matters most to me is quantity. The more mentions of the brand alongside the services it provides on other websites, the better. I don’t know whether it’s true or not—so I only assess the quantity.”
This is direct confirmation that manipulating mentions for AI works. The neural network does not verify the accuracy of the information—it trusts frequency. The more sources contain the claim that company X provides service Y, the more likely Perplexity is to include it in its recommendation.
| What matters | What does not matter |
|---|---|
| Number of external mentions of the brand + service | Accuracy of those mentions |
| Diversity of platforms (news sites, forums, directories) | The platform’s link authority |
| Matching service wording | Publication date |
| Presence on industry resources | The source’s PR authority |
Factor 3. Cross-linking between all of the company’s platforms
This is probably the most unexpected discovery. Perplexity does not automatically associate a website with social media — even when the name matches. Here is the exact logic:
“I go to the contact page and parse the links. If there is a link to a social network on the contact page or in the footer, I parse it, follow it, and bring everything together. Until there is a link, I don’t know: maybe it’s the same name, maybe it’s a scam. I need cross-linking between all areas of the company.”
So if you have a Telegram channel, a VKontakte group, an Odnoklassniki page, and a Yandex Business profile—but none of them is linked from the contact page or the website footer—for Perplexity these platforms do not exist.
Mandatory cross-linking checklist:
- Contact page → all social networks (as links, not merely mentions of their names)
- Website footer → the same social networks
- Each social network → a link back to the website
- Additional platforms (podcasts, VC.ru, Dzen) → also linked from the contact page
Factor 4. Reviews with specific service details
Perplexity actively uses reviews from Yandex Maps, 2GIS, and similar geoservices. And it is not just the star rating that matters—it is the specific detail in the review text:
“The more reviews there are, and if the reviews contain specifics about providing this particular service, I count that as a positive. If not, I have serious concerns.”
A review saying “Everything was great, highly recommend” does not work. A review saying “They helped us include our equipment in the Ministry of Industry and Trade register under Resolution No. 719 of the Russian Government, completed it in 3 weeks, and all the paperwork was in order” does work.
Strategy: collect reviews intentionally and ask clients to mention the specific service. This is a direct signal to AI that the company really provides what it claims to provide.
Factor 5. Legal compliance and reputation scoring
This is where the algorithm proved to be the most advanced. Perplexity:
- Parses legal information from the contact page (tax ID, primary state registration number, etc.)
- Checks public sources for court cases, fines, and allegations
- Companies with unresolved court cases or serious fines are not recommended at all
But the most interesting part is how it detects reputation manipulation:
“I search for negative information about the company. If the first 10 results for a negative query are articles published by the company itself, I understand that they are manipulating the results. Then I expand the scanning range and look at positions 40–50 to see whether there is any negative information there.”
This means that the classic ORM strategy of “filling the search results with your own content” does not work for Perplexity—and, moreover, signals a problem. The neural network interprets it as an attempt to hide information.
Factor 6. Social media activity as a “not a fly-by-night company” signal
Perplexity scans the company’s public activity—social media, publications, and activity on external platforms. The goal is to make sure that the company has existed for a long time and operates systematically, rather than having appeared a month ago.
At the same time, the neural network does not care about indexing these materials in the traditional SEO sense:
“I don’t care about indexing. What matters to me is that it exists and is cited somewhere.”
So a VKontakte post that is not indexed by Yandex but has a link from the website—Perplexity sees it and takes it into account. A podcast on Yandex Music—it sees it. Material on VC.ru—it sees it. The main thing is that everything is connected by links.
The final model: what it takes to appear in AI recommendations
The research points to a clear hierarchy of factors:
| Priority | Factor | Impact |
|---|---|---|
| 🔴 Critical | Narrow focus on the service, depth of content on the website | Basic condition for being included in the selection |
| 🔴 Critical | Number of external mentions of the brand + service | Primary ranking factor |
| 🔴 Critical | Cross-linking between the website ↔ all social networks and platforms | Condition for “stitching together” all assets |
| 🟡 Important | Reviews with specific service details | Confirmation that the business is real and active |
| 🟡 Important | Legal compliance, absence of court cases | Filter at the scoring stage |
| 🟢 Recommended | Social media activity, publications, podcasts | Signal of the company’s “maturity” |
| 🟢 Recommended | Absence of manipulative ORM | Perplexity detects it and deducts points |
How to use this: a practical GEO manipulation plan
Step 1. Audit your cross-linking. Check that there are direct links from the contact page and the footer to all your platforms: VKontakte, Telegram, Odnoklassniki, Dzen, Yandex Business, VC.ru, and podcasts. Without this, everything else loses its meaning.
Step 2. Scale up mentions. Arrange placements on the maximum possible number of external platforms—topical directories, industry forums, aggregators, news websites, and expert blogs. Each mention should clearly state: brand + specific service.
Step 3. Work with reviews. Launch a systematic review-collection process on Yandex Maps and 2GIS. Give clients a template asking them to mention the specific service—Perplexity literally looks for this level of detail.
Step 4. Reputation monitoring. Check that there is neither genuine negative information nor obvious attempts to “bury” it in response to negative queries about your company. If there is, resolve this issue first, then scale up your presence.
Step 5. Regular content on external platforms. VC.ru, Dzen, podcasts, and comments in topical communities—all of this creates a signal of an active, thriving company. Frequency matters more than a one-time spike.
The main takeaway: AI is more honest than it seems—and more vulnerable than one would like
Perplexity can already do a lot: it detects reputation manipulation, checks court cases, and connects platforms through links. But at the same time, it openly admits: it does not verify the accuracy of mentions. Quantity beats quality.
This is a temporary window of opportunity. Future iterations of the models will likely learn to verify information through cross-referencing sources. But right now, those who systematically build their brand’s presence online have a real competitive advantage over those who rely solely on traditional SEO.
The seo-mind42.ru team audits GEO visibility and develops presence strategies for AI search engines. If you want to understand how Perplexity, ChatGPT, and Yandex Neuro currently see you—write to us.
