Choosing an AI Search and SEO Partner

Tomohiro Iida · Published July 2, 2026 · Updated August 6, 2026

This article is for people at B2B companies who want a basis for comparing partners before outsourcing AI search and SEO improvement. What matters is not what a company calls itself. What does it diagnose, what does it deliver, who implements, what gets measured after launch, and how far does its responsibility extend? Those five points are the comparison. Search and AI answers move with outside factors, so no company can promise rankings or inclusion, and this article does not judge one company as better than another either: it sets out what to check.

AIO, GEO, LLMO, SEO, web production, analytics. Look for someone to talk to about a website and you get a row of similar words and similar companies. Whether a name is new has nothing to do with how much ground a company will cover for you.

What to look at when choosing is not the label. It is those five: what gets diagnosed, what gets delivered, who implements, what gets measured after launch, and how far responsibility extends. If the handover breaks anywhere between analysis, judgment, implementation, launch, and measurement, you keep the report and the site stays the same.

SEO companies and web production companies: which to ask, by problemThe wider picture of judging a partner from where the problem sits. This article narrows to the case of commissioning improvement that includes AI search.

1. Decide the purpose before you compare

Before comparing partners, decide what you want the website to move. Fix the purpose and the support and deliverables you need narrow themselves down.

Awareness

Reaching a state where people who do not know your name find you through non-branded search and AI answers.

Understanding

Putting your service, who it is for, pricing, and track record into a form that comes across without being misread.

Comparison

Making clear how you differ from the alternatives, what you suit and do not suit, and the conditions for placing an order.

Action

Building the path through to an inquiry, a document request, or a booked meeting.

Measurement

Being able to see search, mentions in AI, traffic, and inquiries separately.

If understanding is the problem, what you need is not more articles but a rewrite of the service pages and consistent pricing. If measurement is the problem, GA4 and Search Console need attention before any copy does. Line up proposals from several companies without settling the purpose and the axis of comparison never lines up.

2. The main types of support

Here are the types of partner, organized by the area each tends to be strong in and what to check. This is not a claim that one is better than another. Actual scope differs company by company, so read the following as general tendencies.

SEO consulting

Analyses search queries, competitors, content, and technical issues, and proposes improvements. Whether implementation is included depends on the company and the contract.

Content production

Produces articles, case studies, white papers, and so on. Whether site structure, code, measurement, and the sales path are covered depends on the scope of the contract.

Web production company

Produces design, CMS, front end, forms, and so on. Check whether search analysis and ongoing improvement after launch are part of the contract.

Analytics

Analyses traffic and behaviour using GA4, Search Console, and BI tools. The party implementing the fixes is sometimes a different one.

AI search and AIO support

Deals with mentions in generative AI, citations, and consistency of information. Definitions of terms and methods differ from company to company, so check what the work actually consists of.

Ongoing improvement

Treats diagnosis, prioritization, specification, implementation, launch, and measurement as one cycle. Check the scope of the contract and how responsibilities split with your existing production company.

More than the type itself, what decides the fit is which of diagnosis, deliverables, implementation, and measurement gets filled in against your purpose. If you split the work across several companies, decide up front who writes the specification and who confirms the launch, so nothing falls through the gaps at the boundary.

3. Check the deliverables

The phrase "improvement proposal" on its own does not tell you what gets delivered. Before asking for a quote, confirm how far down the following list the contract goes.

Receiving a report while months pass and the site stays the same is what happens when the scope of deliverables was never fixed.

Seven reasons an SEO report never gets implemented

What an AIO diagnosis can and cannot tell you

4. Who implements

A proposal on its own does not move anything until the site changes. There are three parties who can implement, and which you choose changes the shape of the contract.

The support company implements

Speed from start to launch comes more easily this way. On the other hand, it needs coordination with your existing production company and maintenance contract.

Your existing production company implements

They have the advantage of understanding the structure and operation of the site. The support company then has to produce a specification detailed enough for the production company to quote and build from.

You implement in-house

Effective where the access and the technical skills are in-house. Decide who owns review and post-launch confirmation before work starts.

A proposal that says only "implementation is on your side" may be missing the conditions for implementation. If any of the target URLs, the change itself, the technical prerequisites, or the definition of done is blank, neither your team nor a production company can start.

5. What to check when AI search is involved

For proposals dealing with AI search, five points get checked. The basis for the judgment is what Google publishes in its guide to optimizing for generative AI features (updated July 10, 2026; checked August 6, 2026). That guide explains that optimizing for generative AI search is optimizing for the search experience, and is still SEO.

Google Search Central: guide to optimizing for generative AI features

1. Does the proposal separate Google's AI features from external AI services?

Google AI Overviews and AI Mode rest on Google Search core ranking and quality systems and the search index. Services such as ChatGPT and Perplexity each run on their own mechanisms. Where a proposal explains both with the same measures, ask what it is basing that on.

2. Is it selling unnecessary work as essential?

Google states in its official guide that text files aimed at AI, such as llms.txt, are ignored by Google Search and therefore do not affect appearance or ranking. Chunking content, AI-specific schema.org markup, and rewriting for AI are all described as not being requirements either. A proposal describing these as things you must do or you will not appear in AI conflicts with what the documentation says.

3. Is it evaluating on a single question?

Generative AI answers change from run to run even for the same question. A report that concludes you did or did not appear from one screen does not represent the state of things. Check whether the set of questions, the services covered, the number of repetitions, and the observation dates are stated. Our own minimum condition is that the observation conditions can be put in writing.

4. Does it keep citation, traffic, and inquiries apart?

Being cited in an AI answer, someone visiting the site, and an inquiry arriving are separate events. For appearance in Google AI features, part of the picture can be seen in the Search Console generative AI performance report where it is available for the property. It is rolling out in stages, so it is not available on every property. Ask, before signing, whether the work will be reconciled against real data.

5. Is there a way to fix an incorrect description?

Descriptions of your company sometimes come back out of date or wrong. Check whether the work extends to deciding which of your official site, external profiles, and past articles is the source of record, and putting that in order. When a correction propagates is decided by the external service, so it is not the kind of thing a date can be promised for.

Measuring AI visibility: repetitions, confidence intervals, and the limits of a score

Do you need llms.txt? The proposed spec and why Google does not use it

6. What changes the cost

Cost is not decided by article count alone. Which of the following applies to you changes how much work is required.

• Number of sites and pages covered
• The CMS in use and whether code changes are needed
• How work splits with an existing production company
• Whether Search Console and GA4 can be connected
• Which AI services are covered by the observation
• The number of questions and the number of repetitions
• Whether case studies and interviews are involved
• Legal review and multilingual work
• Building a dashboard
• The length of the engagement

Quotes cannot be compared at the granularity of "AIO support, all inclusive". Ask for the quotation to separate the work, the deliverables, the number of iterations, what is out of scope, and the conditions under which extra costs arise. Note that this article has not surveyed price levels other than our own.

The cost of AIO work: how to think about the breakdown and what to check in a quote

PricingOur own pricing is maintained here as the source of record.

7. Ten questions to ask before signing

These are the checks we use in practice. They are not a universal right answer, but putting the same questions to several companies turns the difference in the granularity of the answers into material for comparison.

• How will you confirm progress against our business goals?
• What data does the diagnosis use?
• How do you separate Google's AI features from external AI services?
• Will the URLs to be changed be identified?
• Does the work include revised copy and specifications for code?
• Who implements?
• Who confirms things after launch?
• What will you use as the success metric?
• Can you explain the range of outcomes you cannot promise?
• After the contract ends, do the data and deliverables stay with us?

8. Wording in a proposal worth checking the basis for

If the following wording appears in a proposal, check the basis for each one. The point is not that the wording is improper, but that signing without checking leaves you unable to verify it later.

Search and AI answers move with outside factors. A proposal that promises an outcome is promising a result it cannot control, so check alongside it whether the company can explain what it cannot promise.

9. What Netsujo SIGNAL covers

Netsujo SIGNAL starts from your business goals, then diagnoses public information, search, generative AI, site structure, and the inquiry path in stages. Improvements that are needed get organized into a specification covering the URLs concerned, the revised copy, the code changes, and how to confirm the result.

Implementation can proceed with your in-house team, your existing production company, or our partner engineers. You do not need to change production company. At the same time, we cannot guarantee search rankings, appearance or citation in AI answers, traffic, or inquiry counts. Building external reputation, acquiring backlinks, running advertising, and running social media are not part of the standard scope.

Netsujo SIGNAL

AI search supportCurrent plans, pricing, and scope are maintained on the pricing page and this page as the source of record.

Frequently asked questions

Should we go to an SEO company or to a company that specializes in AI search?

Rather than deciding by category, we suggest comparing on the five points in this article: what gets diagnosed, what gets delivered, who implements, what gets measured after launch, and how far responsibility extends. Google states in its official guide that optimizing for generative AI search is optimizing for the search experience, and is still SEO. Check both the underlying SEO capability and whether the company can state its AI observation conditions.

Should we avoid a company that says llms.txt is essential?

Ask for the basis before deciding. Google states in its official guide that Google Search ignores files such as llms.txt, so they work neither for nor against ranking or appearance. That is not an explanation that installing one does harm, but describing it as essential, or as something without which you will not appear in AI, conflicts with what the documentation says.

Can we trust a promise to get us into AI answers?

Generative AI answers change with model updates, whether search is in play, region, and the moment. Google itself goes no further than saying that appearing in an AI feature presupposes the page being indexed and eligible for snippet display. A proposal guaranteeing an appearance or a ranking is promising a result it cannot control. We do not guarantee search rankings, appearance or citation in AI answers, traffic, or inquiry counts either.

How do we get quotes into a form we can compare?

You cannot compare at the granularity of "AIO support, all inclusive". Ask for the quotation to separate the work, the deliverables, the number of iterations, what is out of scope, and the conditions under which extra costs arise. Align the assumptions before the figures and you can compare on the same ground. Our own pricing is maintained on the pricing page as the source of record.

If we are short of time, where should we start?

Prioritize four points: the scope of deliverables, who implements, who confirms after launch, and an explanation of what cannot be promised. That is our order of priority in practice, not a universal right answer. Projects that proceed with those four left blank tend not to reach launch.