AI SEO Services: What Should Be Included (and What to Skip)

An AI answer about your brand passes through four stages. Each one has a different owner.

• Parametric: what the model knows with no web access, frozen at its training cutoff. Nobody changes this in a quarter.
• Access: whether a crawler may fetch and store your page. You and your CDN control this.
• Retrieval: whether your page enters the model's context for a given question. The engine decides.
• Generation: which retrieved pages the model uses and cites. The model decides.

A critical survey of 45 GEO studies (arXiv, July 2026) concluded that already-retrieved content can change how it gets cited, but no reviewed technique shows a stable, cross-platform effect on being discovered in the first place. Most published GEO effect sizes were measured on pages already sitting inside the model's context.

What that means for buying: the services with evidence behind them work at the access stage, and they work by removing blocks. Ask any provider which stage each line item touches. If the answer is always "generation", you're paying for the part with the weakest evidence.

Crawler access audit

robots.txt rules are per bot, and each bot has a different job. OpenAI's crawler documentation says GPTBot governs training, OAI-SearchBot governs ChatGPT search, and "each setting is independent of the others". So this is valid and common:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

A good audit resolves the rules for OAI-SearchBot, PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot and bingbot, and explains the consequence of each. Then it tests the second gate. OpenAI's help centre says eligibility means allowing OAI-SearchBot and confirming your host or CDN allows traffic from its published IP addresses. A firewall can return 403 to a crawler that robots.txt allows. From 2026-09-15, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, so a site can lose access without anyone editing it. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

Render fixes

Vercel found that ChatGPT and Claude crawlers fetch JavaScript files (11.50% and 23.84% of requests) but don't execute them. Googlebot does render. The service here is a diff: raw HTML against the rendered page, listing every price, product name and claim that only appears after JavaScript runs. The fix is moving that content into server-rendered HTML.

Suppression and eligibility checks

noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers. They often get inherited from a template. Google says a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode. A thorough provider also checks the Search Console generative AI control (including inheritance from a parent property) and your Bing index status in Bing Webmaster Tools.

Accuracy correction

AI platforms get specific facts wrong about specific brands: prices, product names, locations. The service compares what each engine says against your real facts, finds contradictions across your own pages, and states each fact plainly on a page crawlers can fetch. When engines cite other sites instead of yours, being named accurately on those sites is associated with getting into the answer.

Per-engine measurement

Google says AI Overviews and AI Mode "may use different models and techniques". In Otterly's 2026 experiment, Microsoft Copilot produced 2,718 of the 3,934 citations tracked for one site while Google's AI surfaces almost never cited it. A single blended score would have hidden that. Insist on results per engine, with retrieval state disclosed.

These line items appear in many proposals. The primary sources don't support them as ways to get cited.

Two cautions. First, don't overcorrect: Google says existing schema does no harm, so removing it isn't a fix either. Second, IndexNow is fine as free Bing hygiene. It just isn't an AI service.

In your control:

• Which crawlers your robots.txt and firewall let in.
• Whether key content is in the HTML before JavaScript runs.
• Snippet and index controls on your pages.
• The facts stated on your own site, and whether they agree with each other.
• Which third-party listings you correct.

Not in your control:

• What a model learned before its cutoff.
• Whether an engine searches at all for a given question. As of October 2026, ChatGPT search fires at the model's discretion and Claude.ai search is a user toggle.
• Which retrieved pages the model cites. OpenAI's full ranking disclosure is that ChatGPT uses "multiple factors".

This is why a provider who guarantees AI citations is promising something nobody can deliver. A good service promises to find and fix what blocks you, then shows you what changed.

Whether you hire out or do it in-house, this order puts the hard blocks first.

1. Baseline per engine. Record what each engine says about you for real buyer questions. The free checker on this site gives a snapshot for three questions.
2. Access. Resolve robots.txt per bot, then request pages as each crawler to catch firewall or CDN blocks.
3. Rendering. Diff raw against rendered HTML on your pricing, product and about pages. Fix what's missing.
4. Suppression and index. Check snippet directives, the Search Console generative AI control, and Bing index status.
5. Facts. Fix contradictions across your pages. State each key fact in one plain sentence.
6. Third-party sources. Correct wrong details on the sites engines cite for your category.
7. Re-measure. Use the same questions, logged out with memory off, and report each engine separately.

Steps 2 to 4 are binary and checkable. If a provider skips them to start on content, ask why.

• Buying one blended score. Any single AI visibility score averages answers given with and without live search. Ask for per-engine results.
• Treating robots.txt as the whole access story. A firewall can block what robots.txt allows. Ask whether the provider tests both.
• Blanket AI blocks. Blocking every AI bot to stop training also removes you from ChatGPT search. Block GPTBot if you want, but keep OAI-SearchBot separate.
• Trusting headline uplift numbers. The widely quoted "40% more visibility" came from a lab engine with sources loaded into context. Read how to judge GEO case studies with big citation gains.
• Treating prompt tracking as ground truth. ChatGPT rewrites your question into its own search queries. A prompt set is a sample, not the real input.
• Assuming your SEO retainer covers it. Some of the work overlaps, some doesn't. Check whether your current SEO agency already handles AI search before buying a second contract.

Map every line item to a stage and a deliverable. A strong proposal names the check, the evidence, and what you get at the end: a list of blocked crawlers, a render diff, a fact contradiction report, per-engine results.

Questions that sort providers fast:

• Do you test firewall and CDN access per crawler, or only read robots.txt?
• How do you detect content that only appears after JavaScript runs?
• Do you report each engine separately, and say whether search was on?
• Which of your recommendations have a primary source behind them?

For a longer checklist, see questions to ask when choosing an AI SEO agency and the overview of what AI SEO agencies do. If you run a smaller business, start with whether you need an AI search optimization service at all.