Search and AI Visibility
SEO Built on What Google Actually Published
Most SEO proposals you receive in 2026 recommend at least three things that Google has told people in writing do not work. This page names them, cites the source, and explains what we do instead. We run search and marketplace visibility for ecommerce brands and have done since 2019.
Google published a document that voids most AEO advice
In July 2026 Google put out an AI features optimisation guide. Read the sentence it leads with:
"There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."
It goes further and names techniques that do not help. Google lists llms.txt files, chopping content into small chunks, rewriting pages for AI systems, chasing inauthentic mentions, and over focusing on structured data. On llms.txt specifically its wording is blunt: Google Search ignores them.
That matters because llms.txt setup appears on a lot of agency price lists right now. It is a proposal by one developer, first published in 2024, with a second version in August 2026. It has no standards body behind it. Publishers have adopted it. No major search or AI provider has confirmed consuming it.
We will add an llms.txt file if you want one. It costs almost nothing and does no harm. We will not tell you it makes you visible in AI answers, because the company that runs the largest AI answer surface has said in writing that it ignores the file.
FAQ and HowTo rich results no longer exist
This one catches almost everybody, including agencies who genuinely know their craft.
Google restricted FAQ rich results to government and health sites in September 2023. In May 2025 it added a deprecation notice. The feature stopped appearing in Google Search on 7 May 2026, and Google deleted the documentation on 15 June 2026. HowTo went earlier: the rich result stopped showing in 2023 and the docs came down that September. The sitelinks search box markup was retired in October 2024.
So if a proposal promises more search real estate from FAQ schema, the person who wrote it has not checked Google's documentation this year.
FAQ markup is not worthless. It is still valid schema, it still describes your content accurately, and question and answer pairs are genuinely easier for language models to lift. What it no longer does is produce a visible rich result in Google Search. We keep using it for the first reason and never sell it on the second.
Structured data still earns rich results for plenty of types: Product with merchant listing and return policy, Article, Breadcrumb, Event, Job posting, Local business, Recipe, Review snippet, Video and more. For an ecommerce site, Product markup with shipping and return policy is where the real return sits.
What the research actually supports
The paper everyone quotes is GEO: Generative Engine Optimization, presented at a major data mining conference in 2024. It reported visibility gains of up to 40%, and that number now appears on hundreds of agency pages.
A survey of 45 studies published in July 2026 checked the claim. The 40% applies only to documents already retrieved and placed in the model's context window. It is not a measure of whether you get found in the first place. Quoting it as an AI visibility uplift misrepresents the source.
That survey found strong evidence for only two things. Query and document relevance, including position in the context window. And extractable evidence, meaning statistics, clear definitions and direct quotations, which help reuse provided they are factually accurate.
It found something uncomfortable too. One system tested body only optimisation and retrieval dropped by 9%. Changes that help you get quoted once you are retrieved can hurt your chances of being retrieved at all.
A separate ecommerce study from November 2025 tested 15 common listing rewrite heuristics against 7,000 product queries. Ten of the fifteen produced negligible or negative results, and FAQ formatting and authoritative tone were among them. The best simple heuristic moved rank by 0.71 places.
The Core Web Vitals numbers, corrected
First Input Delay has not been a Core Web Vital since March 2024. Interaction to Next Paint replaced it, and the migration deadline passed in September 2024. Any audit still reporting FID is two years stale.
| Metric | Good | Measured at |
|---|---|---|
| Largest Contentful Paint | 2.5 seconds or less | 75th percentile |
| Interaction to Next Paint | 200 milliseconds or less | 75th percentile |
| Cumulative Layout Shift | 0.1 or less | 75th percentile |
Be careful how anyone sells this to you. Google's own page experience document says results in the Core Web Vitals report do not guarantee top rankings, and that Search will still show the most relevant content even when page experience is poor. Speed is a tiebreaker and a conversion lever. It is not a ranking button.
The conversion side is better evidenced. Farfetch A/B tested a 600 millisecond improvement in Largest Contentful Paint on product pages and measured a 1% to 5% conversion uplift. That is a controlled test, not a correlation. Their correlational data separately showed conversion falling about 1.3% for every extra 100 milliseconds of load.
Two things Google shipped in 2026 that change reporting
You can now see AI impressions
Search Console added a generative AI performance report in June 2026, covering AI Overviews and AI Mode. It gives impressions by page, country, date and device.
It gives you impressions only. There is no click data. Anyone presenting AI Overview click numbers from Search Console is presenting something that does not exist. The report is also still rolling out, so not every property has it.
You can opt out without losing snippets
Until June 2026 the only way to stay out of AI Overviews was the nosnippet rule, which also stripped your normal search snippets. Google now provides a dedicated Search Console control with include, exclude and inherit options, and states plainly that it is not used as a ranking or inclusion signal elsewhere in Search.
Most ecommerce brands should stay in. But if you sell subscription content or your value is the answer itself, you now have a clean choice that did not exist a year ago.
Where AI content policy actually stands
Google does not penalise AI written content. The relevant policy is scaled content abuse, and its wording is explicitly method agnostic: creating large amounts of unoriginal content that provides little value, no matter how it is created.
The framework Google publishes for judging content is Who, How and Why. Who made it, how it was made including whether automation is evident to visitors, and why it exists. Google calls the last question perhaps the most important.
On experience, expertise, authoritativeness and trust, Google's own document says trust is the most important and that content does not have to demonstrate all four. That is a more forgiving standard than most agencies describe, and it points the work somewhere specific: named authors, visible dates, real sources, and claims you can defend.
What we do
Technical
Core Web Vitals against field data rather than lab scores, since Google measures the 75th percentile of real visits. Crawl and index coverage, canonical logic on faceted navigation, internal linking, and Product structured data with shipping and return policy where it still earns a rich result.
Content
Pages that answer one question completely, open with the answer, carry a named author and a visible date, and cite sources. That is what the research supports and what Google's trust guidance asks for. It is also what gets quoted by language models, because extractable evidence is one of the two things with strong evidence behind it.
Measurement
Search Console for both classic and generative AI impressions, tracked separately because they behave differently. We report on queries, pages and impressions, and we tell you which numbers do not exist rather than estimating them.
Off page work, stated plainly
Link building is where SEO proposals get vaguest, so here is our position in full.
Google's spam policies name link spam as a violation, and that includes buying links, excessive exchanges, and low quality directory or bookmark submissions. It also names site reputation abuse, which is placing third party content on an established domain to borrow its ranking signals. Guest post farms sit squarely inside both.
What we do instead is narrower and slower. Digital public relations tied to something real, such as original data from your own account, a launch, or a founder with a verifiable track record. Supplier, distributor and trade association listings that would exist whether or not search engines did. Fixing links that already point at broken URLs on your site, which is free and usually overlooked.
We do not sell packages of a fixed number of links per month. If an agency quotes you 40 links for a flat fee, ask where they come from and whether they would still buy them if Google did not exist.
What this has produced
Four accounts we can point at, each with the detail on its own page.
- From zero to $6.2 million. A new brand taken from launch to eight figures of cumulative revenue.
- Advertising cost cut from 35% to 10% in six months. Keyword and placement work, not budget cuts.
- $13,000 in 25 days. A launch window, not a steady state figure.
- A brand built from $5,000. Small starting budget, long horizon.
Since 2019 we have handled more than $100 million in marketplace sales, and our profiles across freelancing platforms carry more than 300 reviews at a 5.0 average.
When you should not hire us
- You want guaranteed rankings. Nobody can promise those. Google publishes no formula and its own documents say good scores do not guarantee position.
- You want AI Overview citations guaranteed. The only peer reviewed survey on this found no technique with a stable, cross platform, causal effect on discoverability.
- Your product pages have no unique content. If you resell a catalogue everyone else lists, fix assortment or merchandising first.
- You need traffic this month. Buy ads. Search work compounds over quarters.
What this costs
Technical and content work starts at $1,000 per month, with a 90 day initial term and 30 days notice after. A one off technical audit is available separately if you would rather take the findings in house. The audit is free with any engagement and yours to keep either way.
Common questions
Does FAQ schema still give rich results?
No. Google stopped showing FAQ rich results in Search on 7 May 2026 and removed the documentation on 15 June 2026. The markup remains valid and still helps language models parse question and answer content, but it no longer produces extra space in Google Search.
Do I need an llms.txt file?
Not for Google. Google's July 2026 guidance lists llms.txt among techniques that do not help and states that Google Search ignores these files. No major AI provider has confirmed consuming it. It is harmless to publish, but it is not a visibility lever.
How do I appear in AI Overviews?
Google says there are no additional requirements and no special optimisations. AI Overviews use retrieval augmented generation on top of the core Search ranking systems, so ordinary search quality work is what feeds them.
What are the Core Web Vitals thresholds?
Largest Contentful Paint at 2.5 seconds or less, Interaction to Next Paint at 200 milliseconds or less, and Cumulative Layout Shift at 0.1 or less, each measured at the 75th percentile of real visits. First Input Delay was retired in 2024.
Does Google penalise AI written content?
No. The policy is scaled content abuse and it applies to low value content produced at scale no matter how it is created. Google's stated test is who made it, how it was made, and why it exists.
Can I see how often I appear in AI Overviews?
Partly. Search Console added a generative AI performance report in June 2026 covering AI Overviews and AI Mode. It shows impressions by page, country, date and device, with no click data, and it is still rolling out to properties.
Can I keep my site out of AI Overviews?
Yes, and without losing normal snippets. Since June 2026 Search Console has a dedicated control with include and exclude options. Google states it is not used as a ranking or inclusion signal elsewhere in Search.
Is the 40% figure from the GEO paper real?
It is real but widely misused. A July 2026 survey of 45 studies found it applies only to documents already retrieved into the model's context, not to whether you get found. One tested approach even reduced retrieval by 9%.
Our Guarantee
What we put in writing
What you always get
- One named account manager who knows your account, not a rotating pool
- A written report every month covering what we changed and why
- Every file, login and asset stays yours, handed over on request
What you never get
- No lock-in. Cancel any month, with no notice period
- No setup fee, and the audit before you start is free
- If we miss the scope we agreed in your first 30 days, that month is free
We do not promise a ranking, a Buy Box share or a sales figure, because those sit with the platform and not with us. We promise the work, the reporting and the terms above, and all three go in the contract.
Get a free technical audit
Core Web Vitals field data, index coverage, structured data that still earns results, and what to stop paying for.
Book a CallSources: Google, AI features optimisation guide. Google, AI features in Search. Search Central updates log. web.dev, Web Vitals. Google, page experience. Farfetch case study. Google spam policies. Search Console generative AI report. Aggarwal et al., GEO, KDD 2024. Martinez, GEO critical survey, 2026.