Author: Vladislav Bobrov, founder of Tier1.shop. In SEO and webmastering since 2009.
Search engine results are ceasing to be a list of links. Yandex accompanies a neural response for already 68% of informational and 43% of commercial queries, and "Search with Alice" is used by 88 million people. Google has a similar picture: AI Overviews appear in almost half of the results and cut organic CTR by 61% (Seer Interactive data: from 1.76% to 0.61%). Hence the panic that "SEO is dead." Short answer: no. What has changed is not "whether links and positions are needed," but what exactly they should provide: citability, brand mentions, and guaranteed indexing. Based on 2025–2026 research, we analyze how neural networks choose sources for answers — and how to get into those sources. Related analyses: how to format an article for citation and is llms.txt needed.
What is happening with traffic: 2025–2026 figures
First, an honest summary of losses — according to data as of mid-2026:
- The share of zero-click (the user received an answer and did not click anywhere) for AI answers is 40–60% versus 8–15% for classic search results.
- According to analyses of Russian-language sites, informational pages lose 30–60% of organic traffic after the appearance of an AI answer for their queries.
- Ahrefs: in the presence of AI Overview, clicks on the first Google position drop by 58%.
And now, what is talked about less often. First, traffic from neural networks is growing rapidly: AI referrals already provide 12–18% of global referral traffic (a year ago it was 5–8%). Second, this is a hot audience: the conversion of transitions from ChatGPT is 7,1%, higher than organic and almost at the level of paid search. The user arrives after the neural network has already "sold" the source to them.
And the main figure that changes the strategy:
Sites quoted in an AI answer receive 35% more organic clicks than those not quoted. The game has shifted: it is not enough to be at the top — you need to be the source that the answer itself refers to.
How neural networks choose whom to quote
A key fact that dispels half of the myths: no major neural network "knows the internet on its own". When answering a question, they all rely on classic search indexes and tops:
| System | Where it gets sources | What this means for you |
|---|---|---|
| ChatGPT Search | Bing Index: according to Seer Interactive, 87% of citations match the Bing top | No site in Bing — no you in ChatGPT |
| Google AI Overviews / Gemini | Google Index (grounding); 38% of citations — from the top-10 of regular search results (Ahrefs) | Classic positions still decide |
| Perplexity | Own PerplexityBot crawler and its own index | The site must be open for crawling by AI bots |
| Yandex Neuro / Alice | Own Yandex top; the share of sources do not from the top-10 fell from 40% to 10% in six months | The neuro-answer cites the organic top more strictly |
Two conclusions follow from this table, which should be accepted as axioms of 2026.
First: an unindexed page does not exist for a neural network. A link, mention, or article — anything placed on a page outside the index, provides exactly zero signal for both classic ranking and AI answers. Indexing has turned from a "hygiene factor" into an entry ticket — and with it, things are so bad for purchased links that we dedicated a separate analysis with figures and a verification guide.
Second: classic SEO has become not less, but more important. AI answers are an add-on to the top. In Yandex, the dependence on the top 10 has only strengthened over the past six months; in Google, more than a third of citations come from the top ten. Everything that moved a site to the top before — quality links, topical relevance, on-page, — now works twice: for positions and for appearing in the answer.
Brand mentions versus "naked" backlinks
The most interesting part of the 2025–2026 research is which external signals correlate most strongly with visibility in AI answers. There is data here based on large samples.
Ahrefs, 75 000 brands. Correlation of brand presence in AI Overviews:
- brand mentions on third-party sites — 0,664;
- branded link anchor text — 0,527;
- branded search demand — 0.392;
- the number of backlinks as such — 0,218.
That is, brand mentions "weigh" approximately three times more than the naked number of links. Neural networks are language models: they learn from texts and cite those who are written about on the web, not those who simply have a link in the footer.
Semrush, 1 000 domains. Here, they looked deeper into the link profile:
- domain authority (Authority Score) correlates with AI mentions at the level of 0,65 — stronger than any quantitative link metric;
- a sensation for link builders: nofollow links for AI visibility work practically on par with dofollow (correlations 0.509 vs 0.504). For a language model, a mention with a nofollow link is the same text as one with dofollow;
- the effect is threshold-based: visibility grows not from "+1 link", but from the profile moving to the next tier of authority.
Put it all together, and you get the portrait of the ideal "AI signal" unit: an indexed thematic page where the brand is mentioned in meaningful text and supported by a link. Not a line in a directory with 200 outbound links, not a rented link in the sidebar of a non-thematic donor, but textual environment + link + index.
I will say without false modesty: we at Tier1.shop arrived at this before the trend. In our donor selection criteria, we have used the metric of domain mentions in Google (уп.G) for years — alongside the number of referring domains and thematic relevance. And the placement from the main page is done within a thematic paragraph — meaning it is simultaneously a link and a textual mention in a related environment. In 2026, this stopped being "our quirk" and became exactly what Ahrefs and Semrush measure.
GEO: how to prepare content that gets cited
A term has already been established for optimization for generative engines — GEO (Generative Engine Optimization). The first academic study — a work by Princeton and colleagues (presented at KDD 2024): the authors ran hundreds of queries through generative engines and measured which changes in content increase the chance of appearing in the answer.
Main conclusions:
- proper GEO optimization provides up to +40% visibility in neural network answers;
- the things that work best are: statistics with source indications, expert quotes, links to primary sources — what makes a text "citable" by design;
- 44% of neural network citations are taken from the first third of the text. Long warming-up introductions are completely dead: the answer and key figures — immediately;
- the effectiveness of techniques varies from niche to niche — there is no universal recipe, you need to test on your own queries.
From practice, this should be supplemented with: clear definitions of terms (neural networks love "what is X — it is..."), comparative tables, FAQ blocks with schema.org markup, visible update dates. By the way, the article you are reading now was assembled according to these same rules — conclusions in the first third, every figure with the name of the study, a table, FAQ. The technique works recursively.
Checklist: what to do in 2026
- Check the indexing of everything that should be cited. Not only in Google and Yandex — set up Bing Webmaster Tools: the Bing index is the gateway to ChatGPT, which is the last thing people in the RuNet remember.
- Check robots.txt for AI bot blocking. GPTBot and OAI-SearchBot (OpenAI), PerplexityBot, Google-Extended. Some hostings and CDNs cut them by default — make sure that closing off from neural networks is your conscious decision, not a default.
- Increase the breadth of mentions. Ten mentions from ten different thematic domains provide more AI visibility than ten links from one. The same principle we gave for links from home pages: distribute across the front.
- Include branded anchor text in the link strategy. The correlation of branded anchor text with AI visibility (0.527) is twice as high as for backlinks in general. A profile consisting only of commercial keys in 2026 loses twice: both in terms of over-spamming risks and in neural network answers. How to compile an anchor list — we analyzed separately.
- Maintain a classic top position for your query cluster. Yandex neural answers are 90% assembled from the top-10, AI Overviews — by more than a third. Link building and on-page have not been canceled — they are now "dual-purpose".
- Write by GEO. Answer in the first third, figures with sources, structure, tables, FAQ.
- Start measuring. AI referral segment in analytics (transitions from chatgpt.com, perplexity.ai, Yandex neural answers) and regular manual checks: whether neural networks cite you for key queries in the niche.
What this means for link strategies
Links are not dead — their KPI has changed. Previously, one thing was required from a link: to pass weight. Now, a good link works toward three goals at once: weight → positions → citability in AI responses, and simultaneously serves as a brand mention — a signal with a correlation of 0,664.
At the same time, requirements for the donor have tightened. The placement page must be in the index — otherwise, it is zero across all three KPIs immediately. The donor must be topical, and the link must be placed within meaningful text, because the language model "reads" the surrounding context, not just the href attribute. Cheap blasts, non-indexed crowd links, and links from junk pages have completely lost value in the AI era: they passed crumbs before, and now they don't even make it into the raw data for neural responses.
Links from home pages in this logic have turned out to be a successful unit of AI signal: the home page is almost always in the index (we guarantee the indexing of every placement), the link is placed in a topical paragraph — meaning it is a link and a mention simultaneously, — and the limit of 5 external links from the home page maintains the concentration of weight. We do not promise inclusion in neural responses — honestly, no one can guarantee that, — but across all measured factors, such placements hit exactly the signals that neural networks take into account.
Frequently Asked Questions
Is SEO dying because of neural networks?
No. All AI responses are built on top of classic search indexes and tops: ChatGPT takes sources from Bing, AI Overviews and Neuro — from their own Google and Yandex tops. To get into a neural network response, you first need to get into the regular top — meaning SEO has become a mandatory condition, not an alternative.
Do links work for getting into AI responses?
Yes, but differently than before. Brand mentions (0,664 according to Ahrefs) and domain authority (0,65 according to Semrush) correlate most strongly with AI visibility, rather than the number of links itself (0,218). The best format is a link within topical text on an indexed page: it works as both a link and a mention.
Do neural networks consider nofollow links?
Practically on par with dofollow: in a Semrush study, the correlations were 0,509 and 0,504. For a language model, text with a nofollow link is the same text.
How to check if neural networks are citing me?
Manually: ask ChatGPT, Perplexity, and Alice 10–20 key questions in your niche and see who they name and who they link to. Systematically: the AI-referral segment in analytics and specialized AI visibility trackers.
Should I open the site to AI bots in robots.txt?
If you want to be present in responses — yes: OAI-SearchBot (ChatGPT Search), PerplexityBot, Google-Extended. Blocking is appropriate only if you consciously do not want to give content to neural networks, and then you must accept the loss of this channel.
How does GEO differ from SEO?
SEO optimizes for the ranking of a list of links, GEO optimizes for getting into the text of the neural network response itself: cited phrasing, statistics with sources, the answer in the first third of the text, structure. In practice, they do not compete: GEO only works for those who have already solved SEO tasks — indexing and positions.
Briefly about the main points
Neural responses take clicks away from classic search results, but are built from them: indexing and top ranking are the entry ticket, brand mentions and authority are the main external signals, and GEO content structure is the multiplier. Links in this system are not dead, but have become more expensive in terms of meaning: each must be indexed, topical, and textual — that is, simultaneously a link and a mention.
Want your link profile to work for both rankings and AI answers — place links from home pages with guaranteed indexing. Questions — to the support bot @tier1shopsupport_bot, we will answer.
