AI SEO Packages: What a Fair Scope Looks Like

A reasonable AI SEO package (also sold as GEO packages or AEO retainers) is mostly technical audit and fix work, plus honest measurement. The scope should read like a list of things the provider will check and repair, not a list of outcomes it will produce.

A fair scope usually includes:

• An access audit of every AI crawler that matters, covering robots.txt and what your firewall or CDN actually returns.
• A rendering check that shows which facts only appear after JavaScript runs.
• An eligibility review for Google and Bing: noindex, nosnippet, max-snippet, data-nosnippet, and canonical conflicts.
• A fact consistency pass so your prices, product names and company details say the same thing on every page.
• Per-engine answer audits that show what each AI platform says about you and where it is wrong.
• A written fix list with owners, so your team or developer can ship the changes.

If a proposal leads with content volume, schema bundles or a proprietary authority score, ask what evidence sits behind each line. The rest of this guide explains why that question matters.

Every AI answer passes through up to four stages. Knowing them tells you which deliverables can work at all.

1. Parametric. What the model knows with no web access, frozen at a training cutoff. Nothing a publisher does this quarter changes it.
2. Access. Whether a crawler may fetch and store your page. This is controlled by you and your CDN.
3. Retrieval. Whether your page enters the model's context for a given question. The engine rewrites the user's question into its own search queries, and you never see them.
4. Generation. Which retrieved pages the model actually uses and cites.

Most published GEO effect sizes were measured at stage four, on pages already inside the model's context. A critical survey of 45 GEO studies (arXiv 2607.14035, July 2026) concluded that no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability. That is the line a package has to respect.

Retrieval is also discretionary. As of October 2026, ChatGPT decides per question whether to search, Google AI Overviews often do not trigger, and Claude's web search is a user toggle in the app. So a single answer tells you little. Repeated, per-engine audits tell you more.

Here is the split a provider should be able to state plainly in its proposal.

The access stage is where a package earns its fee. These fixes are mechanical, checkable, and documented by the engine operators themselves. They work by removing blocks, not by adding signals.

One on-page practice has a positive experimental signal: stating a specific fact (a price, a number, a one-line definition) in HTML that crawlers can fetch gives the engine a sentence it can quote. That is a reasonable deliverable. A promise that it will be quoted is not.

If you already pay an SEO agency, much of this overlaps with technical SEO. Before buying a separate package, check whether your current SEO agency should already handle AI search.

Per-bot crawler access

robots.txt rules apply per bot, and the bots do different jobs. OpenAI's crawler documentation states that GPTBot governs training use, OAI-SearchBot governs ChatGPT search, and "each setting is independent of the others." A site can block training and still appear in ChatGPT search:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

The package should list each retrieval bot (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot, bingbot) and its documented consequence.

Firewall and CDN reachability

A firewall or CDN can return a 403 to a crawler that robots.txt allows. OpenAI's help centre says eligibility needs both robots.txt permission and a host or CDN that allows its published search bot IP addresses. Since 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers. A good package requests your pages as each crawler and reports the status codes.

Rendering

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they can't read client-side rendered content. The deliverable is a diff of raw HTML against the rendered page, listing the prices, names and claims that only appear after hydration. Googlebot renders JavaScript, so this matters most for non-Google engines.

Index and snippet eligibility

Google's AI features documentation says a page must be indexed and eligible to show with a snippet to appear in AI Overviews or AI Mode, and that "there are no additional technical requirements." The package should find stray noindex, nosnippet, max-snippet:0 and data-nosnippet tags, often inherited from a template.

Fact accuracy and off-site presence

The provider should extract your core facts and fix contradictions across your own pages. Where engines cite other sites instead of yours, being named on those sites is associated with appearing in answers. Treat that as an association, not a promised effect.

Walk away from, or strike out, any line that promises one of these:

• Guaranteed citations, mentions or a target score. No vendor controls what the models say.
• An AI rank or position in Google. Google's generative AI report in Search Console shows impressions only. There is no position data to check such a number against.
• "Up to 40% more visibility." That figure comes from a lab engine the GEO paper's authors built, with sources already loaded into context.
• Schema or FAQ schema to get cited. Google says structured data isn't required for generative AI search. Ahrefs' matched test of 1,885 pages found AI Overview citations fell 4.6% after schema was added. FAQ rich results were deprecated in May 2026. Existing markup does no harm, but it is not a lever for AI answers.
• An E-E-A-T score. Google says E-E-A-T isn't a specific ranking factor. There is no score to raise.
• Building domain authority. Domain Authority is a Moz metric that no engine computes. Google does run a news-scoped topic authority system, but it publishes no site-level score.
• An llms.txt file as a visibility fix. Google says Search ignores it, and no other engine documents reading it.
• Chunking or AI-friendly rewriting. Google rejects both and says there is no ideal page length.
• IndexNow pings for faster AI citation. The spec never mentions AI, and a 200 response only confirms receipt.
• Allowing GPTBot so ChatGPT finds you. GPTBot is for training. Search uses OAI-SearchBot.

Share of voice bundles often lean on several of these. Read whether AI share of voice packages are worth buying before you sign one.

Ask any provider to sequence the engagement roughly like this. If they start at step five, ask why.

1. Baseline what each engine says. Run audits per platform for your real buyer questions. The free checker on this site gives a quick snapshot of whether AI engines name you for up to three buyer questions.
2. Fix access. Resolve robots.txt per bot, then test what each crawler actually receives. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.
3. Fix rendering. Move prices, product names and key claims into server-rendered HTML.
4. Clear suppression. Remove unintended noindex and snippet controls, and confirm index status separately in Google Search Console and Bing Webmaster Tools.
5. Fix facts. Make every page state the same price, name and company details, in plain specific sentences.
6. Work on presence elsewhere. Where engines cite review sites, directories or publications, get accurately listed there.
7. Re-audit per engine on a fixed set of questions, and compare against like-sized competitors, not household brands.

Steps two to five are finite. A provider should be able to quote them as a project, with monitoring billed separately.

• Accepting one blended score. Engines behave very differently. In OtterlyAI's 2026 experiment, Microsoft Copilot kept citing two AI-generated sites after Google deindexed both, while Google's AI surfaces almost never cited them. A single number would have hidden that.
• Treating prompt tracking as ground truth. Engines rewrite your question into their own queries. A tracker's prompt list is a sample, not the real input.
• Paying for content volume. The same Otterly test saw Google remove both sites of roughly 1,000 AI-generated posts each, with no manual action in Search Console. Google's spam policies call this scaled content abuse.
• Ignoring hidden-text offers. Some providers pitch injected instructions or content cloaked for AI crawlers. Research shows manipulated content can sway AI search agents, which is exactly why it is a reputational and policy risk. Never buy it.
• Skipping the pricing questions. Ask what is a one-off fix and what is recurring. This guide to fair, transparent pricing for AEO tools covers what to ask.

If you are hiring a person rather than a package, the checks in hiring an AI SEO consultant apply. Agencies reselling someone else's service should also review what to check in white-label AI SEO.

How long should an AI SEO package run? The fix work is usually finite. Access, rendering, suppression and fact consistency can be scoped as a project with a clear end. Monitoring is the recurring part, because AI answers change as engines update models and retrieval. A fair proposal separates the two, so you are not paying a monthly retainer for fixes that were finished in the first few weeks.

Can a provider guarantee my brand appears in ChatGPT? No. A provider can make your site eligible: OAI-SearchBot allowed, your firewall passing it, and your key facts in server-rendered HTML. Whether ChatGPT searches for a given question, retrieves your page, and cites it is decided by the engine. OpenAI's own description of ranking says placement is not guaranteed. Treat any guarantee as a red flag.

Should the package include schema markup? It can include schema as general hygiene, but not as a way to get cited. Google says structured data isn't required for generative AI search, and Ahrefs found AI Overview citations fell 4.6% on pages after schema was added. Existing markup does no harm, so removing it is not needed either. Just don't pay a premium for it as an AI lever.

How should results be reported? Per engine, per question set, with dates. Reports should say whether answers came with live search or without it, list wrong facts found, and compare you with brands of similar size. Avoid reports that show an AI rank or position, since no engine publishes position data for generative answers. Presence, accuracy and citations, disclosed by platform, are the honest units.

What is the difference between GEO packages and AI SEO packages? Mostly the label. GEO, AEO and AI SEO packages are sold for the same goal: appearing accurately in AI answers. Judge them by scope, not name. The useful deliverables are the same in each: crawler access, rendering, index and snippet eligibility, fact consistency and per-engine audits. The promises to reject are also the same.