The short answer.
AI SEO services are either SEO produced with the help of AI tools, or optimisation aimed at AI search engines. The first changes how the work is made. The second changes where you are trying to appear. Many sales pages blur the two, so the first job for a buyer is to find out which one is on the table.
Neither is a trick. The problem is paying for one and expecting the other: faster article production will not, by itself, change what ChatGPT says when a buyer asks for a shortlist.
This guide is mostly about the second meaning, where scopes are vaguest. It is also the kind we offer: our GEO, AEO and AI SEO service starts from the same dated baseline of buyer questions.
Meaning one: SEO done with AI tools.
In this meaning, the service is ordinary search engine work and AI is the production method. The aim is still to rank in normal search results. Language models and automation do parts of the job faster: grouping keywords, drafting briefs and first drafts, writing titles and descriptions, suggesting internal links, and sorting through crawl data.
Google has no objection to the method. Its guidance on generative AI content says: “Generative AI can be particularly useful when researching a topic, and to add structure to original content.” What it objects to is volume without value. Its spam policies list, as an example of scaled content abuse, “using generative AI tools or other similar tools to generate many pages without adding value for users”. The same guidance says it is “critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.”
So an AI-assisted SEO service is only as good as the people checking the output. Look for four things:
- A named reviewer. Someone who knows your subject reads every page before it goes live.
- Your facts, not the model’s. Claims about your company come from your documents.
- Fewer, better pages. The plan starts from buyer questions, not a page count.
- Honest framing. The provider says where AI is used.
If those four are present, the “AI” in the name is a detail about how the provider works.
Meaning two: optimisation for AI search.
In this meaning, the service aims at the answers AI engines give, not only at a position in a list of links. The work is to make your company easy for those engines to find, read, describe accurately and name when a buyer asks a question in your category. You will also see it sold as AI search optimisation services, GEO or AEO; our explainer on how AEO, GEO and SEO differ covers the vocabulary. Our guide to generative engine optimisation covers the research behind the term.
Buyers ask for it because research habits have changed. G2, the software review platform, reported on 6 October 2025, from its August survey of more than 1,000 B2B software buyers, that half said they now start the buying journey in an AI chatbot instead of Google Search. That is a vendor survey. Independent data shows search behaviour changing too. Pew Research Center’s panel of 900 US adults, tracked in March 2025, clicked a search result about half as often when an AI summary was shown.
How much of it is really new?
Less than the sales pages suggest, and more than nothing. Google’s guide to its generative AI features says plainly that “the best practices for SEO continue to be relevant”, and its page on AI features states there are “no additional requirements to appear in AI Overviews or AI Mode”. For Google, good SEO is the entry ticket.
Three things sit outside ordinary SEO:
- Other engines, other crawlers. ChatGPT and Perplexity run their own crawlers and use their own mix of sources.
- What other websites say about you. Engines compare providers using review platforms, directories, lists and publications.
- Measuring answers. A rank tracker cannot tell you whether you were named in an answer.
Our guide to how ChatGPT decides which companies to recommend covers the mechanics.
Which one are you being offered?
Read the deliverables, not the headline. If every line is about producing pages or reports faster, it is the first meaning. If the lines name buyer questions, AI engines, crawlers, third-party sites and answer tracking, it is the second.
| What to compare | SEO done with AI tools | Optimisation for AI search |
|---|---|---|
| Typical line items | AI-assisted research, briefs, drafts, metadata, automated audits | Answer baseline, crawler access, page restructuring, off-site sources, answer tracking |
| What gets measured | Rankings, organic visits, enquiries | How often you are named or cited in answers, AI referral visits, enquiries |
| Main risk | Many thin pages nobody checked | Activity you cannot verify, or promises about placement |
| First question to ask | Who reviews each page before it goes live? | Which buyer questions, in which engines, and can we see the answers? |
Buyer, in Perplexity: “Which payroll providers suit a UK company with 80 staff, salary sacrifice schemes and no in-house HR team?”
What the first kind of service does about it: nothing directly. It may publish more payroll articles, more quickly.
What the second kind does: records who is named for that question today, checks the engine’s crawler can read the provider’s service page, makes sure the page states company sizes and schemes supported in plain text, and looks at which review sites and lists the answer drew on.
An invented question, written to show the difference. It is not a real query or answer.
A useful test: ask the provider to mark each line of the proposal “with AI” or “for AI search”. One who cannot has probably renamed an existing SEO package.
The five parts of a credible service.
A credible service for AI search contains five parts: an audit, technical access, content structure, off-site sources and measurement. Leave one out and the others are weakened.
| Part | What is done | Ordinary SEO or new? |
|---|---|---|
| 1. Audit | Your buyers’ questions are run through named AI engines; answers and sources are recorded | New. A rank tracker cannot show who was named in an answer |
| 2. Technical access | Crawler rules, firewalls, rendering and indexing are checked | Mostly ordinary SEO; the addition is the other engines’ crawlers |
| 3. Content structure | Key pages are written or reshaped to answer buyer questions directly, with evidence | Ordinary SEO done well, aimed at buyer questions |
| 4. Off-site sources | Profiles, reviews, directories and lists that engines read are corrected or earned | Partly new: review sites, directories and lists, not link counts |
| 5. Measurement | The same questions are re-run on a schedule; AI referral visits and enquiries are tracked | New: repeated answer tests, read beside your analytics |
Audit and technical access.
The audit shows what AI engines say about your category today, and the technical work makes sure they can read you at all. They come first because they shape the content plan. We describe what a technical SEO engagement covers in a separate guide.
Part 1: the audit
A proper audit starts from questions, not keywords. The provider agrees the questions your buyers ask before they shortlist, runs each in the engines they use, and records who was named, which pages were cited and what was said about you. It should be dated and state the country tested from. You can do a first version yourself with our method for checking your AI search visibility. Or request the free AI visibility report, which runs a sample of your buyers’ questions and shows who is named.
Part 2: technical access
Each engine documents its own crawler. OpenAI says OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features”, and that sites opted out of it “will not be shown in ChatGPT search answers”. Perplexity says PerplexityBot “is designed to surface and link websites in search results on Perplexity” and recommends allowing it in robots.txt. An old rule against scrapers, or a firewall that challenges unknown bots, can remove you from those answers unnoticed.
For Google, the list is familiar. Its AI features page says SEO fundamentals still apply, including “ensuring that crawling is allowed in robots.txt, and by any CDN or hosting infrastructure”. Our guide to AI crawlers, robots.txt and structured data covers the checks in detail.
Content structure and off-site sources.
Content work makes your own pages easy to quote; off-site work shapes what other websites say about you. Engines use both when they compare providers, so a service that only writes blog articles covers half the picture.
Part 3: content structure
The pages that matter most are usually service pages, comparison pages and case studies, not blog posts. The work is to make each one answer a buyer’s question in its opening lines, in plain text, with dated facts. Our guide to the page patterns AI engines quote shows the structures that get lifted and credited.
You do not need to rewrite everything into fragments. Google’s guide says: “There’s no requirement to break your content into tiny pieces for AI to better understand it.” For harder-to-explain products, see our piece on search content for complex B2B products.
Part 4: off-site sources
Engines lean on what the wider web says about a brand. Ahrefs, an SEO software company, compared 75,000 brands in May 2025 and found that mentions of a brand on other websites tracked its presence in Google’s AI Overviews more closely than its backlink count did. The author stressed that correlation is not causation.
In practice this means accurate profiles on review platforms and directories, genuine reviews from real customers, and a place on the lists and publications that cover your category. Our guide to the off-site sources AI engines use for B2B firms ranks them by effort and value, and our study of where AI answers get their sources shows the mix we recorded.
There is a wrong way to do it. Google’s guide warns that “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem”. The US Federal Trade Commission’s rule, announced on 14 August 2024, bans fake reviews and testimonials, including AI-generated ones, and paying for reviews on condition of a particular sentiment. A provider who offers to “supply” mentions or reviews is offering you a liability.
Measurement.
Measurement means re-running the audit’s questions on a schedule and connecting the result to visits and enquiries in your own analytics. Without it, nobody can tell whether the other four parts worked.
There are now first-party reports to draw on. Google announced generative AI performance reports for Search Console on 3 June 2026, showing how often your URLs appeared in generative AI features in Search and Discover; a later note says they reached all websites by 31 August 2026. The post lists impressions as the only metric. Microsoft opened an AI Performance report in Bing Webmaster Tools as a public preview on 10 February 2026, showing citations in Copilot and Bing’s AI summaries, and the phrases the AI used when it retrieved your pages.
Neither tells you whether you were named in ChatGPT or Perplexity, so answer testing is still needed, and it has to be repeated, because answers vary. In a study by SparkToro and Gumshoe.ai, reported by Search Engine Journal on 30 January 2026, ChatGPT and Google’s AI each returned the same list of brands in fewer than one in 100 repeated runs of a prompt. One run of one question is an anecdote. The share of runs in which you are named is a measurement.
What a report should let you do
- See every question, engine, date and answer behind any headline number.
- Compare this month with the baseline, on the same questions.
- See AI referral visits in your own analytics, and the enquiries that followed.
Tools can do the repetition, within limits. Google’s guide says: “No third-party tool has access to our internal ranking or AI systems.” Our comparison of AI visibility tracking tools explains what each one measures, and our guide to reporting AI visibility to leadership shows how to present the numbers.
What to ignore.
Ignore any line that sells a file, a tag, a promise or a page count as the service itself. Each item below is sold in AI SEO packages, and each is undercut by a source cited in this guide.
| Line item | Why you can discount it | Ask for instead |
|---|---|---|
| An llms.txt file as a headline deliverable | Google says you do not need “new machine readable files, AI text files, markup, or Markdown” for its Search | Proof that AI search crawlers can fetch your key pages |
| “AI schema” or special markup | Google says there is no special schema.org structured data to add for its AI features | Structured data that matches the visible page |
| A promised placement or date in AI answers | Answers change from run to run; nobody outside the engine controls them | The share of repeated runs that name you, against a baseline |
| Hundreds of AI-written pages a month | Google’s spam policies name mass pages without added value as scaled content abuse | A short list of pages, each tied to a buyer question |
| Mentions or reviews “supplied” for you | Google calls inauthentic mentions unhelpful; the US rule bans fake and bought reviews | Named sites, real customers, disclosed outreach |
| A proprietary score with “inside” data | Google says no third-party tool has access to its ranking or AI systems | The questions, dates and answers behind the score |
When these items make up most of a proposal, the five parts that do the work are missing.
Google’s own advice on hiring an SEO says no one can promise a first-place ranking on Google. AI answers vary from run to run, so a promise about them deserves less weight still.
The service checker.
Open a proposal or a service page and tick every line it clearly contains. The checker tells you which kind of AI SEO service it is, how many of the five parts it covers, and which lines to question. Nothing you tick is sent anywhere.
Tick the lines the proposal contains.
AI search parts covered: 0 of 5 · AI-tool lines: 0 · Lines to question: 0
The checker does not score quality: a proposal can list all five parts and do them badly. It gives you the right label for what you are reading, and questions to send back.
How to scope it.
Scope an AI SEO service by fixing the questions first, the engines second and the deliverables third. Discuss the fee only after those three are written down, so every scope can be compared with any other.
- 01 · ASKList 15 to 25 questions buyers ask
- 02 · ENGINESName the engines and the country
- 03 · BASELINERecord today’s answers and sources
- 04 · GAPSPick the parts the baseline calls for
- 05 · RE-TESTAgree the schedule and the report
Finish steps one to three before any longer commitment.
Match the scope to your starting point
- You are missing from most answers. Start with technical access and off-site sources. New articles will not help if crawlers cannot read you or no third-party site describes you.
- You are named, but described wrongly or thinly. Start with your service and proof pages, and correct the facts on your main profiles.
- You are named and described well. Keep measuring, and spend on the path from visit to enquiry.
- Your ordinary SEO is weak. Fix that first; Google’s AI features draw on pages eligible in normal search.
Packages, projects and retainers
Fixed AI SEO packages suit the first meaning better than the second. Work on AI answers is hard to package, because the right mix of the five parts depends on what the baseline finds. Start with a baseline. The choice between an audit, a project and a retainer, and what the market charges, is covered in our guide to what AI search optimisation costs. Our own service page shows how we scope AI search work.
Who does the work
Your own team knows the buyer questions and the facts, and your developer can often fix crawler access quickly. Answer testing, off-site work and page restructuring are where an outside AI SEO agency tends to add most. Our list of questions to ask an SEO or AI search agency will help you test theirs. We set out what a GEO or AEO agency does week by week in a separate guide.
Questions buyers ask.
What are AI SEO services?
The term covers two things. One is ordinary SEO produced with AI tools for research, drafting and audits. The other is optimisation for AI search engines such as ChatGPT, Perplexity and Google’s AI Overviews, so that your company is found, cited and named in their answers.
Is AI SEO different from normal SEO?
Partly. Google says the best practices for SEO continue to be relevant for its AI features. The additional work is access for other engines’ crawlers, what third-party sites say about you, and measuring answers instead of rankings.
What should an AI SEO service include?
For AI search, five parts: a dated audit of what AI engines say in answer to your buyers’ questions, checks that their crawlers can read your pages, restructured key pages, work on the third-party sites engines read, and re-testing on a schedule.
Are fixed AI SEO packages worth buying?
For AI-assisted content production, a package can be fine if a named person reviews every page. For work on AI answers, start with a baseline instead, because the scope depends on what it finds.
Sources: Google Search Central (generative AI features guide; AI features and your website; generative AI content guidance; spam policies; do you need an SEO?; generative AI performance reports, 3 Jun 2026); Bing Webmaster blog, 10 Feb 2026; OpenAI crawlers; Perplexity crawlers; Pew Research Center, 22 Jul 2025; G2, 6 Oct 2025; Ahrefs, 26 May 2025; Search Engine Journal, 30 Jan 2026; US FTC, 14 Aug 2024. All accessed 12 November 2026.