SEO still matters in the age of AI search. Google's generative answers rest on the same ranking systems as the ordinary results page, and a page only reaches them if it is indexed and allowed to be shown with a snippet. What changes is not the need for optimisation but what counts as a result: alongside a position in the listing there is now a mention of your site inside an answer.
Key takeaways
The format of the answer changed, not the way sources are chosen. Google says so plainly: there are no extra requirements for appearing in AI features, a page has to be indexed and not blocked from snippets, and the same robots.txt, nosnippet and max-snippet directives govern it as in ordinary search (Google Search Central documentation).
What people do on the results page, however, really is shifting, and independent measurements show it.
The conclusion behind these numbers is not that search optimisation is over but that part of the attention now stays inside the answer. So a second job is added to the familiar one: getting into that answer, which we cover in our article on optimising for generative search.
Optimisation has four blocks: technical foundations, structure, content and metadata. Choosing keywords belongs to one of them; it is not the whole job.
Technical requirements are the condition without which nothing else counts: the page has to open, be served to the crawler and land in the index correctly.
Core Web Vitals are the three measures the Chrome team uses for page experience: rendering of the main block, responsiveness to interactions and layout stability. They are read at the 75th percentile of real visits, on phones and desktops alike.
| Metric | What it measures | Good threshold |
|---|---|---|
| LCP | The moment the largest visible element finishes rendering | up to 2.5 seconds |
| INP | The delay before the page responds to an interaction | up to 200 ms |
| CLS | Layout movement during loading | up to 0.1 |
INP replaced the earlier FID metric and became the main one in 2024; the definitions and thresholds are published by the Chrome team (web.dev). This is a measure of how comfortable the page is, not a separate lever for positions: a slow site gets in the way of people and crawlers alike, while a fast one guarantees nothing on its own.
Site structure answers one question: which query each page exists for. Two pages answering the same question compete with each other; a question no page answers goes to a competitor.
A keyword map is useful because it turns guesses about demand into a list of pages. Broad queries define sections, while narrower ones become individual articles for people who already know what they want. Which sections a company needs at the start is covered in our article on why a business needs a corporate website.
It helps to split queries by intent straight away: learn, compare, buy. A page built for “what is” and a page built for “order” are written differently and lead the reader to different actions, and trying to merge them on one screen usually solves neither job.
Content is judged on one thing: does the page close the question or send the reader off to keep looking. Everything else follows from that.
The easiest way to check completeness is the list of questions a person asks straight after the main one.
Tip Write the page description as an invitation to click, not as a keyword store. If search decides that text from the page describes it more accurately, it will show its own version and your description will never be seen.
No. Google builds the snippet automatically from the content of the page and uses the meta description only when it gives the reader a more accurate idea of the page (Google Search Central help). The description influences whether the link gets clicked, not where the link appears.
The title works differently. Google generates the title link automatically from the title element, the visible heading and other signals, and may substitute its own version when it finds a problem: an empty title, the same title across every page, or one unrelated to the content (Google Search Central help).
The practical rule is simple: the title is written for the content of the page and for the query, the description for the person choosing between five links.
Pages reach the results page, but the quality of the site counts too. Google puts it without euphemism: search assumes that an individual page, including a new one, matches the quality of the other pages on the domain (Google spam policies).
What follows is usually learned the expensive way: ten strong pages plus two hundred empty ones is not ten chances, it is a blurred signal. Sections built only to exist are better left unpublished or kept out of the index.
The second practical conclusion concerns internal links. A page with no link pointing to it from inside the site looks accidental to a crawler: it takes longer to reach and its place in the structure is harder to read. A link with a descriptive anchor from a neighbouring article works better than one more item in the menu.
Foundations first, visibility in AI answers second, because the second job rests on the first: a page that is not indexed and not shown with a snippet will not reach an AI answer either. The reverse does not hold — good foundations alone guarantee no mentions.
| Question | Search optimisation | Visibility in AI answers |
|---|---|---|
| What it competes for | The position of a page in the results | A brand mention and a link inside the answer |
| Main lever | Indexing, structure, match to the query | Crawler access, completeness, citability |
| Where results show | Search Console, Yandex Webmaster | Platform reports and checks on a fixed query set |
| What is counted | Impressions, clicks, positions | Mention rate, share of voice, citation rate |
| How it builds up | Gradually, as the index grows | Depends on how often the platform recrawls the site |
The right-hand column is unpacked in our article on optimising for generative search. How to bring both jobs into one monthly measurement is shown in the article on the AI visibility report.
The local market differs in two things: two search engines and three languages on one site. Neither makes the work fundamentally harder, but both add steps that are easy to forget.
The language of the queries deserves a separate look. People search for the same product in Russian and in Uzbek, and demand across the two rarely matches: in some niches Russian dominates, in others Uzbek, and in a few the Latin script mixes with Cyrillic. Guessing here is expensive — it is cheaper to collect queries for each language separately and see where the demand actually sits.
For a Russian-speaking audience one more layer has appeared: AI answers in Yandex search are produced by Alisa, which has its own sources. What to do about it is covered in the neighbouring article on generative search.
The mistakes repeat from project to project, and almost all of them are visible before the budget is spent.
Syntra Systems does not sell promotion as a separate service: basic search optimisation is part of building the site. We map queries and section structure at the design stage and run the basic setup on handover — a company website from $1,900, an online store with basic SEO and GEO setup from $10,000. The scope is described on the website development page.
After that the site has to be maintained: indexing, speed, errors and content all need watching. That work sits inside project development, from $1,500 a month, with a fixed number of team hours, sprints and a quarterly review of priorities (technical support). We promise neither positions nor deadlines: search does not work that way, and a promise like that is checked a month later.
Let’s discuss your project
Tell us what you need, and we will estimate the timeline and cost and suggest a solution.
No. AI services answer from pages they were able to fetch and read, and in Google Search the generative features rest on the same ranking systems. What shifts is how attention is distributed inside the results, not the need to keep a site readable and reachable.
Two different timelines. We plan and deliver our own work — indexing, structure, speed and content — in stages, but how long a search engine takes to notice it depends on niche competition and the site's age, so we don't promise rankings by a date. Instead we record impressions, clicks and positions as a "before" baseline and track the trend in Search Console and Yandex Webmaster.
You need the wording people use to describe their problem, and it lands in the text by itself when the page honestly answers the question. Repeating the query in every paragraph is harmful: the page reads worse and answers worse.
Content — a fast page with no answer to the question is useless, and speed is a usability score, not a separate ranking lever on its own. The thresholds still have to be met on real visits, phones included: the largest element on the page renders within 2.5 seconds, response to an action within 200ms, and the layout barely shifts while loading.
No. The order of results is set by algorithms and depends on the query, device, language and location of the person; paid placement in organic results does not exist. What can be guaranteed is the scope of work and a transparent before-and-after measurement.
The core work is shared: indexing, structure, content. The differences are in the details — its own webmaster console, its own sitemaps and its own business listing. Building a separate site for the second search engine is not necessary.
Sources
