SEO and AI Search Improvement: A Practice Log

Tomohiro Iida · Published June 25, 2026 · Updated July 7, 2026

This is a practice log of SEO and AI search improvement work we carried out on our own site, netsujo.jp. It is a full correction of a practice log we published previously. The earlier version described some measures as effective that a later Google official guide (updated June 29, 2026) stated were "not needed" or came with "no special additional requirement." Since this is a practice log, things we judged correct at the time we wrote them can later turn out to be wrong. To keep that history intact, we don’t delete what we did in the past — we append our current assessment as a correction.

netsujo.jp is a young site with still very small traffic. We haven’t had dramatic results, and we won’t claim results we haven’t had. The value of this article isn’t a success story — it’s laying out, honestly, "what we did," "what we observed," and "what we can and can’t claim as cause," each kept separate.

One premise for the whole article, stated up front: Google’s official guide states that optimizing for generative AI search is optimizing for the search experience — in other words, SEO itself. There is no special extra requirement to appear in AI search; the condition is that a page is indexed and eligible for snippet display. Given that premise, much of what the earlier version treated as "AI-specific measures" needed its positioning reconsidered.

Key takeaways

This is a practice log of SEO and AI search improvement work on our own site, netsujo.jp, and a full correction of a previously published practice log. What we did in the past isn’t deleted; our current assessment is appended as a correction.
Google’s official guide (updated June 29, 2026) states that optimizing for generative AI search is optimizing for the search experience — SEO itself. There’s no special extra requirement for AI search; the condition is that a page is indexed and eligible for snippet display.
For each measure we separate date implemented, what we did, what we observed, and our current assessment, and disclose honestly both what we can’t claim as cause and measures with no confirmed effect. The value of this article is in laying that out honestly, not in a success story.

Measures we implemented, and later observation and correction

For each measure we record four points: date implemented, what we did, what we observed, and our current assessment (including any correction). "Date implemented" is when we carried out the work ourselves; the Google official update date (e.g., June 29, 2026) is when official guidance was updated — we keep the two distinct rather than conflating them.

Measure	Date implemented	What we did / observed	Current assessment (including corrections)
1. Structured data (JSON-LD)	June 2026, site-wide	We built out Organization, WebSite, Article, BreadcrumbList, FAQPage, and other schemas in JSON-LD, assigned by page type, and cross-referenced organization, person, and site via @id. The structured data validated as correct in Google’s Rich Results Test, but we have not observed a causal link between adding it and being cited in AI answers.	The earlier version implied structured data was a lever for AI visibility — we correct that here. Official Google guidance states structured data is not required for generative AI search and that there is no special schema.org markup to add. It remains useful for page understanding and rich-result eligibility, but we can’t claim it as cause for AI citation. (FAQPage specifically is corrected separately in measure 5 below.)
2. Installing llms.txt / llms-full.txt	June 2026	We placed an AI-facing summary file, llms.txt, and a more detailed llms-full.txt, at the site root, summarizing our company overview, key pages, and contact pathway. After installing it, we have observed no change in Google Search display or traffic that can be attributed to it.	This was the earlier version’s biggest error — it treated llms.txt as an effective measure, and we correct that here. Google’s official guide (updated June 29, 2026) states there’s no need to create a new machine-readable file, AI text file, markup, or Markdown to appear in Google Search. llms.txt is not listed as a ranking factor in official guidance, and we could not confirm any effect on Google Search after installing it. It’s not harmful, but we correct our earlier claim that "installing it gets you into AI." (Also listed again below under measures with no confirmed effect.)
3. Entity alignment (consistent naming)	June 2026	We standardized our company name, address, and representative’s name across every page and file, and registered with and cross-linked to external knowledge bases. Inconsistencies within the site were resolved. Whether this raised AI’s recognition accuracy can’t be verified in isolation.	Consistent naming is a basic hygiene item that helps readability for both people and crawlers, and is worth continuing on its own merits. But we can’t prove the claim that "doing entity alignment gets you correctly represented in AI." We treat it as part of the baseline, not a standalone piece of magic — correcting our earlier framing.
4. Answer-first (conclusion-first) structure	June 2026	We restructured each article to lead with a summary and write conclusion-first. Readability for readers improved. We haven’t observed a causal link to whether AI cites the content.	The earlier version presented this as "writing for AI citation." Official Google guidance states there’s no need to write in a specific style purely for generative AI search. Leading with the conclusion is genuinely useful for readers, so we continue doing it, but we correct the framing of it as an "AI-specific must-do technique."
5. Hub-and-spoke content structure	June 2026	We built a structure linking a central hub article to related spoke articles via internal links. Readers can now navigate between related articles. The scale is too small to isolate any effect on ranking or traffic.	Internal-link design has long been effective baseline SEO, and that positioning doesn’t change. We continue it as ordinary SEO practice, not as an AI-specific measure.
6. Allowing AI crawlers	June 2026	We allowed major AI crawlers to crawl the site via robots directives. Crawling itself became possible. We have not observed a causal link between allowing crawlers and being cited.	The ordering relationship — you can’t be a citation candidate if crawling isn’t allowed — still holds. But this is a precondition, not a sufficient condition that "allowing it gets you cited." We also keep, unchanged from the earlier version, that allowing crawlers carries a trade-off: your content may be used in AI training and answers.
7. Improving titles and descriptions (CTR)	June 2026	We identified pages with impressions but low clicks, and rewrote titles and descriptions to align with search intent. After the update, we have not observed a clear improvement in click-through rate; some pages remain flat.	Title changes take time to be reflected and re-evaluated by search engines, and the effect tends to be slow to show. At this point it’s closer to a measure with no confirmed effect, and we list it again under that heading below. We continue to observe it.
8. Measurement (GSC, GA4, baseline)	June 2026, baseline recorded just before the work began	We recorded a baseline for indexing status and traffic on each page before starting, and have since maintained ongoing measurement under the same conditions via GSC and GA4. Newly published pages have gotten indexed after the work (confirmed via GSC URL Inspection) — a verifiable fact. Effects on ranking formation, traffic, and inquiries have not yet materialized, given the small scale.	This measure needs no correction. The backbone of a practice log is exactly this: continuing to observe. Claiming what worked as cause requires running this ongoing measurement for a long time.

Start from a free web diagnosis to check how your site is currently described to AI and search.

See Netsujo SIGNAL plans

What we can’t claim as cause (separating hypothesis from fact)

As the record above repeats, for most measures we cannot prove the causal claim that "doing this got us cited by AI" or "raised our ranking." Keeping that distinction unambiguous is the whole point of this article.

What we can state as fact: The date a measure was implemented, what we did, and the results we could observe (for example: a newly published page got indexed — a verifiable fact).
What remains a hypothesis: An early, qualitative sense of things going well (a feeling that "our name seems to be coming up more in AI answers") is not causal data — it’s still a hypothesis. The sample is small, and AI answers vary by model, search feature, region, and time, so one or two observations can’t establish causation.
The premise Google’s official guidance sets: Official guidance states there is no special requirement to appear in generative AI search beyond being indexed and eligible for snippet display. A direct causal link between individual measures and AI visibility is not guaranteed even officially.

To be able to claim causation, we’d need to set an observation period, measure repeatedly under the same conditions, and build up a larger sample. We’re partway through that process.

What Is AIO, GEO, and LLMO?

Measures with no confirmed effect

Alongside successes, we also honestly log the measures where we could not confirm an effect.

llms.txt’s effect on Google Search: We installed it, but could not confirm any effect on Google Search display or traffic. Google’s official guidance (updated June 29, 2026) states plainly that there’s no need to create AI text files, and does not list it as a ranking factor.
FAQ structured data producing a rich result: Google ended the display of FAQ rich results in search listings in May 2026. Keeping FAQ structured data in place is not a problem, and Google still uses it for page understanding, but you can no longer expect it to appear as a rich result in search listings. The earlier version described FAQPage as "a measure with a display effect" — we correct that here.
Title and description improvements raising CTR: Click-through rate on the pages we rewrote has stayed flat, and we haven’t confirmed an improvement at this point. Because reflection and re-evaluation take time, we don’t call this a failure — we keep it under ongoing observation.

Hiding these would turn a practice log into marketing copy. A record of what didn’t work is exactly what feeds the next decision.

A checklist for working from facts

Items to check if you want to do the same thing on a factual basis for your own site.

Are you recording "date implemented / what we did / what we observed / assessment" separately for each measure?
Are you keeping "the date you implemented something" and "the date Google’s official guidance was updated" from being conflated?
Are you distinguishing, in writing, "what you observed as fact" from "a feeling or hypothesis"?
For AI visibility, are you avoiding claiming a causal link to any single measure? (Official Google guidance states special schema, special writing styles, and AI text files are not required.)
Are you keeping FAQ structured data without expecting a rich-result display effect from it? (Display ended in May 2026.)
Is baseline SEO in place — crawling, indexing, snippet eligibility, internal links, mobile display?
Are you recording measures with no confirmed effect too, rather than hiding them?
Have you set an observation period and are you repeating measurement under the same conditions?

What we deleted or corrected from the earlier version

A list of what we deleted or corrected from the earlier version in this full rewrite. As a practice log, we also keep a record of what changed and how.

Earlier version	Correction
Presented llms.txt as an effective measure	Corrected — official guidance states it’s not needed, and we couldn’t confirm an effect ourselves (see sections above)
Implied FAQPage produced a rich-result display effect	Corrected — display ended in May 2026 (see sections above)
Implied a causal link between structured data and AI visibility	Corrected — the causal link is unconfirmed (see sections above)
Presented early qualitative impressions as results	Now labeled explicitly as hypothesis
Listed only successful measures	Added measures with no confirmed effect
Mixed together the date we implemented something and the date Google’s official guidance was updated	Now explicitly distinguished throughout

The point of this practice log isn’t to tell a success story — it’s to lay out, separately and honestly, what we did, what we observed, and what we can and can’t claim as cause. As Google’s official guide (updated June 29, 2026) states, optimizing for generative AI search is optimizing for the search experience itself — SEO — with no special extra requirement. That’s exactly why we correct measures whose standing changed later, like llms.txt and FAQ rich results, and keep the measures with no confirmed effect on record rather than hiding them. Claiming causation requires setting an observation period and repeating measurement under the same conditions. We’re partway through that process.

Frequently asked questions

Should I install llms.txt?: There’s no need to install it expecting an effect on Google Search. Google’s official guide (updated June 29, 2026) states plainly there’s no need to create AI text files to appear in Google Search. llms.txt is not listed as a ranking factor in official guidance, and we ourselves could not confirm an effect after installing it. It’s not harmful, but the grounds for spending limited effort on it are thin.
Is it still worth adding FAQ structured data?: You can no longer expect it to produce a rich-result display effect in search listings — Google ended FAQ rich-result display in May 2026. Keeping the structured data in place is not a problem, and Google still uses it for page understanding, but our current assessment doesn’t treat it as "a measure that increases display."
Does adding structured data get you cited by AI?: That causal link hasn’t been proven. Official Google guidance also states structured data is not required for generative AI search, with no special schema.org markup that needs adding. Structured data remains useful for page understanding and rich-result eligibility, but we can’t claim it as cause for AI citation.
Why does this article have so few concrete results?: netsujo.jp is a young site with still very small traffic, and we haven’t accumulated results we could claim as cause. We don’t write that we achieved results we haven’t achieved. We record verifiable facts (such as a newly published page getting indexed) separately from impressions that remain a hypothesis, and we’re continuing to observe.