Hiring an AI SEO Consultant or Specialist: What They Should Deliver

An AI SEO consultant, sometimes called an AI SEO specialist or GEO consultant, is paid to explain why AI engines do or don't mention your brand. The honest version of that job produces three things.

• An access diagnosis. Can each AI crawler fetch your pages, read the content that matters, and show it in answers?
• Accuracy findings. What do ChatGPT, Gemini, Perplexity, Claude and Google's AI features say about you, and which specific facts are wrong?
• A measurement plan. How will you track change per engine, and which numbers are real measurements versus the consultant's own model?

Everything else is optional. If a proposal skips these and opens with content production, schema or a proprietary score, it's selling activity rather than diagnosis. The same test applies if you're comparing firms rather than individuals: see how to vet an AI SEO agency.

A consultant who can't explain this mechanism can't tell you what to fix. Every AI answer passes through four stages.

1. Parametric. What the model knows with no web access, frozen at a training cutoff. Nobody changes this quickly.
2. Access. Whether a crawler may fetch and store your page. You and your CDN control this.
3. Retrieval. Whether your page enters the model's context for a given question. The engine decides, per query.
4. Generation. Which retrieved pages get used and cited. The model decides.

The research matters here. A July 2026 critical survey of 45 GEO studies (arXiv 2607.14035) found that already-retrieved content can change its citation, but no reviewed technique showed a stable, cross-platform causal effect on discoverability. Most published GEO effect sizes were measured on pages already inside the model's context. In plain terms: the tactics most often sold work, at best, on stage four, and only after stages two and three already went your way.

Retrieval is also optional. As of October 2026, ChatGPT searches the web at the model's discretion, and Claude.ai search is a user toggle. So any single visibility number mixes answers given with and without live web access.

Be clear on this before you sign anything. It sets what success can reasonably mean.

In your control

• Which crawlers your robots.txt allows, bot by bot.
• Whether your firewall or CDN lets those crawlers through.
• Whether key facts (prices, product names, locations) appear in the HTML before JavaScript runs.
• Whether noindex, nosnippet, max-snippet:0 or data-nosnippet hide content you want shown.
• Whether your own pages state your facts consistently and specifically.

Not in anyone's control

• What the model learned in training.
• Which queries the engine actually runs. OpenAI says ChatGPT search rewrites a prompt into one or more targeted queries.
• Which retrieved pages get cited.
• How prominent your brand already is. Brand size is strongly associated with citation, and it isn't a lever.

One route sits in between. When engines cite other sites instead of yours, being named on those sites is associated with appearing in the answer. A consultant can help you find those sites. They can't promise the effect. That's why agencies that guarantee AI citations deserve suspicion.

This is the highest-value part of the engagement because the evidence is strongest here. Ask for findings on each of these.

Per-bot robots.txt. OpenAI documents separate crawlers, and its bots page says each setting is independent. Blocking GPTBot stops training use. OAI-SearchBot controls whether you appear in ChatGPT search answers. A blanket AI block often removes search eligibility by accident. A correct split looks like this:

User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

Network reachability. A firewall or CDN can block a crawler that robots.txt allows. OpenAI's help centre says eligibility needs both crawler permission and a host or CDN that allows its search bot's IP addresses. From 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, while Search stays allowed. A site can lose access without anyone editing a file. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

Rendering. Vercel's crawler study found ChatGPT and Claude crawlers fetched JavaScript files (11.50% and 23.84% of requests) but didn't execute them. Googlebot renders; most AI crawlers don't. The consultant should diff raw HTML against the rendered page and list facts that exist only after JavaScript runs.

Index and snippet eligibility. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link in AI Overviews or AI Mode, with no additional requirements. Check Search Console and Bing Webmaster Tools separately. Also check whether the Search Console generative AI control has been switched off, or inherited from a parent property, where that control is available.

Founders rarely hire a consultant because AI ignores them. They hire one because AI says something wrong: an old price, a discontinued product, a competitor's feature.

Good accuracy findings are specific. Each one should state:

• The engine and whether it searched the web for that answer.
• The question asked.
• The exact wrong claim.
• The correct fact and where it lives on your site.
• The likely cause, such as two pages stating different prices, or the fact appearing only after JavaScript loads.

The fix is usually on your own site. Stating a specific fact (a price, a number, a one-line definition) on a page crawlers can fetch gives the engine a sentence it can quote. That's the only positive experimental signal for on-page content in the evidence. A consultant should also check whether your site contradicts itself, because the pricing page, about page and footer often disagree.

If an answer came from a model with search off, no page edit reaches it until the next model generation. A good consultant tells you which findings are fixable now and which are not.

Measurement is where weak consultants hide. Ask how every number in the report is produced.

A sound plan:

• Reports each engine separately. Google says AI Overviews and AI Mode may use different models and techniques. A blended score hides where you're strong and where you're absent.
• Records retrieval state. Was web search on for each answer? If the tool can't say, the number mixes two different things.
• Uses first-party data where it exists. As of October 2026, Search Console's generative AI report shows impressions only, with no clicks or position. Bing Webmaster Tools' AI Performance report shows citations and the grounding queries Microsoft's AI actually ran. Neither has an API for that data.
• Labels models as models. Any composite score is the vendor's construct, not a measurement.

For a quick baseline before you hire anyone, the free checker on this site shows whether AI engines name you for up to three buyer questions. It's a snapshot, which is useful for framing the conversation. If a consultant proposes a share-of-voice dashboard, read whether share of voice packages are worth buying first.

Hold the consultant to this sequence. Each step depends on the one before it.

1. Baseline. Capture what each engine says about you today, per engine, with retrieval state noted.
2. Access. Fix robots.txt per bot, firewall and CDN blocks, rendering gaps and accidental snippet suppression. These are binary and checkable.
3. Index status. Confirm Google and Bing index the pages that matter. One being healthy tells you nothing about the other.
4. Facts. Correct contradictions on your own site and state key facts plainly in fetchable HTML.
5. Third-party presence. Identify the sites engines cite for your buyer questions, and work on being accurately named there.
6. Re-measure. Compare against the baseline, engine by engine, and report what changed without claiming causes the data can't support.

Notice what's missing: mass content production. Google's spam policies treat generating many pages without adding value as scaled content abuse, whatever tool wrote them. For scoping and pricing structure, see what a fair AI SEO package includes.

These pitches are common, and primary sources contradict each one.

A balanced note: existing schema does no harm. A consultant who tells you to rip it out is overcorrecting.

Other mistakes buyers make:

• Trusting big before-and-after charts. Ask how the study controlled for brand size and retrieval state. More on judging GEO case studies.
• Accepting prompt-based tracking as ground truth. Engines rewrite queries, so a fixed prompt list is a sample of a proxy.
• Skipping interview questions. Bring a list. These questions to ask an AI SEO agency work for solo consultants too.
• Buying content before access is fixed. Content behind a 403 or a JavaScript wall doesn't exist to the crawler.