SEO and AI: What Has Changed Recently
Over the past few weeks, a lot of interesting developments have emerged in SEO, AI search, content, and digital PR.
Google is integrating AI more and more actively into familiar products, AI Overviews are taking up a growing share of commercial search results, and the first formal standards are already beginning to appear around GEO.
We have collected seven news stories that are definitely worth paying attention to.
Google is starting to summarize your emails with AI
Google is expanding AI Inbox in Workspace: Gemini will be able to automatically generate summaries of incoming emails. For PR, this is a fairly noticeable change — now a pitch needs to be clear not only to the journalist, but also to the algorithm that may potentially show them a shortened version of the email.
BuzzStream and Aira tested how Gemini summarizes PR pitches. The average summary was around 29 words long, which means most of the original email simply did not make it into the final summary.
At the same time, 87.1% of key-information matches came from the first half of the email, while 81% came from the first two paragraphs.
There is an interesting pattern with bulleted lists as well. Information from them appeared in 65% of summaries, but the first bullet point was included in 69% of cases, while the second appeared in only 15%.
The algorithm almost never carried over links: they appeared in approximately 1% of summaries. Expert quotes appeared in 18%.
There is one more important point. Google says that AI Inbox may take the sender’s interaction history into account. According to BuzzStream, journalists who had already replied to a particular contact later opened their emails around 1.5 times more often and replied 19 times more often.
So the familiar recommendation to “put the most important information at the beginning of the email” becomes even more relevant. The main news hook, the key number, and the reason to pay attention to the pitch should be clear literally from the first lines.
31 AI articles — and the site’s organic performance started falling apart
JumpFly shared a fairly telling client case in which the team decided to scale content with the help of AI.
Before the experiment, the site already had more than 500 manually written pieces of content. The team then published 31 AI-assisted articles. Each piece was additionally reviewed and refined by a subject-matter expert for around 90 minutes.
The first such materials started being published in July 2025.
After the December 2025 Core Update, blog clicks over a 28-day period dropped by 30.7%. By May 2026, Google had already failed to index the eight most recent posts for more than two months.
The situation continued to get worse.
By August, the blog had lost 67.8% of its clicks, while its share of the site’s total traffic dropped from 45.4% to 24%. The decline also affected other parts of the project: the homepage lost 77.2% of clicks year over year.
In the end, JumpFly removed all AI-assisted content and returned to its previous editorial model.
At the same time, this case should not be interpreted as proof that Google penalizes AI content. Google does not prohibit the use of AI as a tool.
The problem is rather different: automation makes it possible to scale content very quickly, but if the search engine considers that content insufficiently valuable, the problem scales together with the volume.
JumpFly also warns that recovery after quality issues or spam signals may take from 6 to 18 months or longer.
Small businesses underestimate Google more than their customers do
BrightLocal surveyed 1,002 local business owners in the US and found a fairly large gap between how businesses perceive Google and how consumers actually use it.
Only 22% of business owners named Google as their main source of new customers. Word of mouth ranked higher at 35%, as did social media at 33%.
The picture is different for customers.
61% start their local search on Google: 52% use regular search results, while another 9% use Google Maps. At some point during the decision-making process, 71% of consumers use Google.
85% of surveyed businesses have a Google Business Profile, but only 35% actively manage it.
30% had updated their profile within the past week, and 72% had done so at least once during the past three months. At the same time, 21% of businesses had not updated their profile for more than three months.
There is an interesting connection with AI search here as well.
In a separate study, BrightLocal identified 415 directories and platforms that LLMs used as sources of information about local businesses. Notable sources included Google Business Profile, Yelp, Facebook, and TripAdvisor.
This means up-to-date local information is no longer needed only for traditional search results. It may also influence what information AI systems receive about a company.
Local SEO is probably not ready for retirement just yet.
AI Overviews are appearing more and more often in commercial queries
Over the course of 12 months, Semrush tracked a fixed sample of 400 commercial queries across banking, e-commerce, and software.
The study covered five types of commercial queries:
- alternatives;
- brand vs brand;
- best X for Y;
- pricing/cost;
- reviews.
Over the year, the share of queries with AI Overviews increased from approximately 45.75% to 88%.
The growth affected every category. By August 2026, AI Overviews were appearing in roughly 82-96% of queries depending on the type.
One of the most noticeable jumps occurred for reviews: the share of AIO increased from around 30% to almost 90%.
For best X for Y queries, the figure reached 96.12%.
E-commerce search results also changed significantly: the share of commercial queries featuring AI Overviews grew from 32.62% to 85.11%.
This is an important shift.
AI Overviews are increasingly becoming more than just an informational-query feature. They are actively appearing in situations where users compare products, read reviews, choose a service, or are already fairly close to making a purchase.
As a result, commercial projects now have to think not only about their position in the traditional TOP results, but also about which sources and brands Google displays inside the AI block.
There are calls to stop evaluating AI search only by clicks
Mike King from iPullRank has suggested looking at AI search not as another performance channel, but rather as an upper-funnel branding and PR channel.
The reason is clear: if AI is evaluated only by traffic, the numbers still look fairly modest.
In one study involving 100 websites, ChatGPT sent approximately 0.62% of the traffic volume generated by Google organic search.
In another study, organic search generated around 39 times more traffic and roughly 30 times more conversions than AI assistants.
Another analysis — 632 tests across 12 websites — found no stable relationship between growth in AI visibility and subsequent increases in branded search or direct traffic.
King does not dispute these numbers. He suggests interpreting the channel differently.
A user may see a brand in a ChatGPT response or another AI system, remember it, return to it later, or make a decision without clicking a link directly.
In that case, measuring the impact only by the last click is genuinely difficult.
For the upper funnel, King suggests looking at share of voice in important prompts, mention frequency, citation position, and sentiment.
Branded search, direct traffic, and self-reported attribution can then be used as later-stage indicators — more on a quarterly horizon than a monthly one.
It is still too early to say that the market has arrived at a single system for measuring AI search. But the discussion is gradually shifting from “how many clicks did we get?” to “what position does the brand occupy inside AI responses?”
China has introduced a separate rulebook for GEO
Following a scandal involving the promotion of a fake product through AI-generated answers, China introduced a voluntary GEO standard.
The document consists of five parts and attempts to define where normal optimization ends and manipulation begins.
The appearance of such a standard is interesting in itself. GEO is still a fairly young and poorly standardized field: different companies use different tools, measurement methodologies, and even different definitions of AI visibility.
The Chinese document attempts to formalize this process.
For example, the standard introduces requirements for measurement methodology and a minimum of 30 prompts per measurement. It also describes source requirements and specific prohibited practices.
Service providers may use their own methodology, but the client should understand exactly what is being measured and how the metrics are calculated.
Special attention is also given to guarantees.
A practical analysis of the standard suggests documenting in advance which results the provider cannot guarantee, conducting repeated measurements, and avoiding conclusions based on a single response from an AI system.
In essence, this is one of the first cases in which GEO is being formalized as a full-fledged service with its own methodology, metrics, limitations, and responsibilities.
That makes this experiment interesting even for companies that do not operate in the Chinese market at all.
Source — Search Engine Journal
The same query in ChatGPT can be built from different data every time
Seer conducted a small experiment to see how ChatGPT searches for information within the same query.
The team ran the same hashed prompt several times and saved not only the final answers, but also the search queries and URLs the system used internally.
The results differed noticeably.
On September 3, ChatGPT found 14 unique URLs. On September 16, it found 40.
However, the number of pages that made it into the final citations changed only slightly: from 5 to 6.
This means that in the second case, 34 out of 40 discovered pages did not appear among the citations visible to the user at all.
The September 17 experiment was even more interesting.
Two runs of the same prompt around eight minutes apart did not produce a single fully identical internal search query.
The sample is small, so it is too early to draw broad conclusions about how ChatGPT works.
But there is an important point here for measuring AI visibility.
A page can enter retrieval — meaning the system found and considered it — but still not appear in the final answer. That is very different from a situation where ChatGPT did not see the page at all.
This means that evaluating only final citations does not show the full picture.
For more stable measurements, it makes more sense to run priority prompts several times, keep the wording identical, and look not only at the final response but also at how consistently the same sources appear.
A single AI visibility measurement is better treated as one sample rather than a final result.
What does this all mean?
The main common trend across all these news stories is the same: familiar SEO metrics are gradually becoming insufficient to show the full picture.
AI Overviews are taking up more and more commercial search results. AI systems collect information differently even for identical queries. Local data is starting to be used not only by traditional search, while GEO is gradually developing its own rules and measurement methodologies.
At the same time, traditional SEO is not disappearing. On the contrary, it is becoming even more important to understand where search and AI systems get their information from, which pages they consider high quality, and what a brand’s presence looks like across multiple channels at once.
© PhoenixProject, with full or partial copying of the material, a link to the source is required.
Comments: 0