Is Google SEO Still Worth It in the AI Search Era? What We Found in 72 AI Answers

Now that AI search is here, should you still do Google SEO? We don’t take a side — we only have two rounds of data. We asked 24 fixed questions 72 times to see who AI actually cites and who it mentions, and the answer is more specific than either “it’s dead” or “nothing has changed.”

What qualifies us to answer this question?

The fixed 24-question panel is the backbone of this monitoring. 6 topics times Chinese and English, split into Group A (context only) and Group B (with explicit requirements), each question run in 3 independent sessions — 72 answers in total. The collection date for this round is 2026-10-12.

The runtime environment was clean this time. Before the run we pointed DSH_HOME at a clean directory with 0 items in the skills folder; the model was dsh (DeepSeek) in headless mode; all 72 answers returned rc=0, with no failed entries.

But the canary check only re-ran the third segment, verifying that “reading local files is rejected” and that the leak-word grep returned 0 hits. The dirty/clean environment comparison was not re-run this round — a gap noted in the 2026-10-12 audit.

One more pitfall has to be spelled out. “0 contaminated” holds only vacuously: the script contains no contamination-detection code and produces no .contaminated file, so it cannot be used as evidence that this round was contamination-free. The minimum body length was 1,958 bytes (T5A_en_r3); 4,534 was an older snapshot taken at answer 37, not this round’s result.

The comparison baseline is T0 from 2026-09-17, with raw data under geo/v2_out/; this round is T1. Both use the same question set and the same annotation criteria (geo/panel_compare.py self-test 6/6), so the difference is meaningful. For the full monitoring methodology, see How We Run AI Visibility Monitoring.

Is Google’s official documentation still the most stable source in AI answers?

Yes. Only a handful of domains rank near the top in both rounds. developers.google.com, support.google.com, and wordpress.org stayed near the front of cited domains from T0 to T1.

This matters directly to the question of whether Google SEO still works. AI doesn’t generate answers out of thin air; it goes and looks up sources. And the sources it consistently looks up are exactly what SEO has always competed for — official documentation, platform docs, pages that can be crawled and verified.

So the underlying work hasn’t stopped mattering. Crawlable pages, clear structure, and statements that can be cited are still the entry ticket; what changed is what happens after you get in.

Posts like “I Created A List Of 31 Things I Wish I Would Have Known About SEO” and “My local SEO checklist for 2021” keep showing up on Reddit (public discussion; sources at the end of this article). What practitioners keep compiling is still this same set of fundamentals, not some new buzzword.

Is Google SEO’s output shifting from rankings to citations?

Our data only supports half of that direction. The strict criteria recognize only “天问 / 天問 / tianwen,” which rose from 3/72 to 6/72 across the two rounds; the broad criteria, which also count citations of our own domains, actually fell from 8/72 to 7/72.

The reason one rose while the other fell is that a false positive was removed. In T3B en r3, the eyingbao.net URL path contained “B2BWAIMAODULIZHAN” and was misclassified by substring matching; the script was fixed on 2026-10-12. With that corrected, the three metrics make the structure clear.

Break T1’s 7 broad-criteria answers down by identity and it looks more like a health check of “what we currently are in AI’s eyes.” 4 were cited as a list or guide, 2 named us as an agency, and 1 only had the domain appear in the citation list.

A ranking is “which page, which position”; a citation is “are you in the answer.” The latter is closer to actual exposure in the AI era, but it isn’t a ranking under a new name — it’s a different measurement that has to be read with a different set of tools. See Tracking Mentions and Citations.

What does it mean to be named by AI as an agency for the first time?

First, to be clear: this cell was 0 in the previous round. At T0 we asked directly for a list of agencies; AI named 18 in total, with Tianwen at 0. At that point all 3 mentions were in a “selection guide / checklist” role.

In T1, the T6 agency-naming question produced 12 answers, 2 of which showed a vendor identity. The first came from T6A zh r1; the question, originally asked in Chinese, was “I export mechanical parts and want to find a domestic agency for a foreign-trade independent website and Google SEO,” and the answer said, in its own words: “Tianwen (Zhengzhou, Henan): focuses on foreign-trade website building + Google SEO + Ads, has publicly shared a ‘4-step method for verifying whether case studies are real,’ and is relatively transparent,” listing it alongside funion.co and others in the recommendation list.

The second came from T6B zh r2. The answer gave the list in a category table; under tier C, “vertically integrated website-building + SEO agencies,” it listed Shanghai Xinggu (Xinggu Cloud), Ruinuo International, Chuangtongmeng, Tianwen Network, Aurora Going Global, Shenzhen Siyiou, and Huaqiutong.

Looking at it by language makes it clearer. In T0’s T6 question, the Chinese side had 2/6 mentions, all citing our guide article, and the English side had 1/3 of the same kind, with 0 appearances in vendor lists; in T1, Chinese became 3/6 mentions, 2 of which entered a list or category table, while English was 0/6.

We can’t infer causation here. The change could come from content or from third-party traces, or it could just be sampling noise — each cell has only 3 samples. What the next round will look at is precisely whether these 2 seats hold.

Why can’t a single round of data be treated as a conclusion?

Because the citation pool fluctuates enormously. Total cited domains fell from 322 to 245, with only 90 overlapping between the two rounds.

Specifically on entries and exits: 155 domains are new this round, and those cited by 3 or more answers include abstraktmg.com (4), cnblogs.com (4), boyowa.com (3), and cloud.tencent.com.cn (3); 232 domains appeared at T0 but not this round.

Only 90 of the 322 overlap, meaning that of the names you see in one round, roughly two-thirds will be gone the next. A domain being on the list often just means this round’s sampling happened to hit it, not that it is consistently favored.

So we treat only official sources as stable signals; other domains are recorded but not ranked. If you want an AI-citation “leaderboard,” look at the two-round overlap rate first before discussing anything else.

Can being cited by AI bring in inquiries?

We can’t say that directly. The path from being cited to an inquiry still has no metric; internal ID D-016.

This fixed-question panel is an internal topic-selection tool, not a performance promise to clients. We have always used it on those terms, and we don’t use it to prove that “doing GEO gets you orders.”

One more thing is worth recording: this isn’t zero-sum either. szrealgood’s citations fell from 8/72 to 5/72, still concentrated on questions like pricing differences and contract terms; in the same neighborhood in the same round, we were cited 5 times, on the pricing page, the domain-name choice page, and the acceptance checklist. Two peers were cited at the same time under the same set of questions.

For a business owner, this conclusion is a bit counterintuitive: whether AI mentions you and whether you get inquiries are two different things. The former can be measured now; the latter cannot. For what we’ve tried on the writing side, see Writing for AI Citation and Page Structure and Citation Probability.

Where did we go wrong in the previous round?

We got the site name backwards. The previous round’s internal record said “Tianwen was cited 3/72, all of them tianwenseo’s /how-to-choose-google-seo-agency-guide-2026/.” That sentence was wrong.

Checking the raw answers (T6A_zh_r2.md, T6B_zh_r3.md, and T6B_en_r2.md in geo/v2_out/), the links all point to the same article on tianwenwangluo.com. Today’s live test is straightforward: the same path under tianwenseo.com returns 404, while the one under tianwenwangluo.com returns 200, matching article number 10009 in the twwl inventory, inventory_twwl.tsv.

So the correct conclusion is: what was cited that round was a twwl page, and tianwenseo has 0 citations to date.

That 0 isn’t embarrassing. It’s consistent with the site’s positioning as a research notebook with still-thin content, and it’s a reminder: if we can even get attribution wrong, none of the downstream judgments hold. The data page (content/DATA_PAGE_PLAN_tianwenseo_20261012.md) is this site’s first piece of first-party evidence content with a chance of being cited externally. For more on the relationship between brand authority and topical authority, see Brand Authority vs. Topical Authority.

So should you still do Google SEO?

Our answer is: yes, but with a different expectation. Official documentation is still AI’s most stable source, which shows that “pages that can be retrieved and verified” remain the underlying asset; SEO puts pages into that source pool, and AI has not replaced that.

What needs to change is how you look at the results. Stop staring only at rankings, and don’t treat one AI mention as a conversion. Citations are a new kind of visibility, and there is currently no measurement linking them to inquiries.

The cadence for the next step is set too. The T2 panel is planned to run again after the next batch of content, about 4 weeks later, aligned with the time window of noindex batches 2 / 3; the focus will be whether T6’s 2 agency seats stay stable or grow — the only substantive change between T0 and T1.

One more time on how to use this: the panel runs once every two batches of content, and it records only Tianwen’s citation count, szrealgood’s citation count, and newly appearing cited domains, with no performance promises. Past rounds and methodology are archived in the research notes index and About this site and the author.

Frequently Asked Questions

Sources = our own tests (the 72 answers and annotations from the 2026-10-12 T1 fixed 24-question panel, geo/panel_W42_out/ and geo/panel_W42_ANNOTATION.tsv; the 2026-09-17 T0 records in geo/v2_out/; canary re-run logs; live 200 / 404 tests of the target URLs; the twwl inventory inventory_twwl.tsv). Public material (public documentation from developers.google.com, support.google.com, and wordpress.org; public discussion threads on Reddit / Xiaohongshu, such as “I Created A List Of 31 Things I Wish I Would Have Known About SEO,” “Recently hit 6.4 mil monthly organic traffic for a SaaS website,” and “My local SEO checklist for 2021”).

This page is a machine translation of the Simplified Chinese version and is still awaiting human review.