A 30/60/90 Day AI Visibility Plan: Access, Accuracy, Then Reporting

Most GEO roadmaps open with content tactics. The evidence points the other way. A July 2026 critical survey of 45 GEO studies (arXiv:2607.14035) found that no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability. The levers that hold up are access levers, and they work by removing blocks.

So this quarter follows the order in which an engine meets your site:

• Access: can the crawler fetch and store the page at all? You control this.
• Retrieval: does the page enter the model's context for a given question? The engine decides, partly based on what it can reach.
• Generation: which retrieved pages get used and cited? The model decides.
• Parametric: what the model knows with no web access. It is frozen at a training cutoff, and nothing you publish this quarter reaches it.

You fully control the first stage. You influence which facts the later stages can quote. You control nothing else. That is why measuring before fixing wastes a month, as covered in why crawler access comes before tracking.

Month one finds and removes the blocks. Work through these steps in order.

1. Read robots.txt bot by bot. Rules are per bot and independent. OpenAI's crawler documentation says GPTBot governs training use, while sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Check OAI-SearchBot, PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot and bingbot. The most common error is a blanket AI block that removes search eligibility along with training. A site can allow search and refuse training:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

2. Test the network layer. A firewall or CDN can block a crawler that robots.txt allows. OpenAI's help documentation says eligibility requires allowing OAI-SearchBot and letting its published IP addresses through your host or CDN. Request a page as each crawler and look for 403 errors, 429 rate limits or challenge pages. As of October 2026, Cloudflare blocks its Training and Agent crawler categories by default on ad-displaying pages for new domains, new sites on existing accounts and existing free-tier customers, a change dated 15 September 2026. The AI Crawler Checker requests a page as each AI crawler and shows which ones the firewall turns away.

3. Diff raw HTML against the rendered page. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. Googlebot renders. Most AI crawlers don't. Anything that appears only after JavaScript runs (pricing tables, tabs, "load more" content) may be invisible to them. Move the facts that matter into server-rendered HTML.

4. Find accidental suppression. Search templates for noindex, nosnippet, max-snippet:0 and data-nosnippet. Google's AI features documentation 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. If Search Console shows you the generative AI inclusion control, confirm it isn't opted out, including by inheritance from a parent property.

5. Record a baseline. Write a fixed set of buyer questions and run them on each engine, logged out with memory off. For every answer, save the text and note whether it showed citations. The free checker on this site gives you a snapshot for up to three buyer questions. Keep the question set unchanged for the whole quarter.

AI platforms get specific facts wrong about specific brands. Month two finds where yours go wrong and fixes what you can.

1. Build one fact sheet. List prices, plan names, product names, founding year, headquarters, integrations and ownership. One owner signs it off.
2. Find contradictions on your own site. Compare the pricing page, about page, footer, old blog posts, press pages and any markup. An old price on a 2023 post is a fact an engine can still retrieve.
3. Fix at the source. Update or retire every conflicting page. State each fact as a plain, specific sentence in HTML a crawler can fetch. A stated fact gives an engine a sentence it can quote, which is the only positive experimental signal for on-page content. It is not a guarantee.
4. Compare the baseline against the fact sheet. Log every wrong fact by engine, question and whether the answer was grounded.
5. Sort the errors. A wrong fact in a grounded answer usually traces to a page the engine retrieved, yours or someone else's, so you can find and fix it. A wrong fact in an ungrounded answer comes from training data and will not move this quarter.

What to leave out: rewriting pages "for AI", chunking content, adding FAQ schema or publishing llms.txt. Google's AI optimization guidance says none of these are needed for its generative features, and controlled tests of citation-focused rewrites found them ineffective or harmful. Existing markup does no harm, so don't rip it out. It just isn't a lever. The playbook on retiring unsupported GEO tasks shows how to take them off the roadmap.

When engines cite other sites instead of yours, being named accurately on those sites is the route into the answer. The evidence shows a strong association here, not a promised effect.

Third-party listings

1. From your grounded baseline answers, list the domains cited for your buyer questions.
2. Sort them into three groups: sites you can edit (directory profiles, marketplace listings, partner pages), sites where you can request a correction, and sites you can't touch.
3. Bring every listing you control into line with the fact sheet.
4. Check index status per engine. Open Google Search Console and Bing Webmaster Tools separately. Bing's index backs Microsoft Copilot, so a healthy Google profile tells you nothing about Copilot.

Some vendors sell hidden text, prompt injection or content cloaked for AI crawlers. Research on LLM search agents shows manipulated pages can sway answers once retrieved, which is exactly why engines treat it as abuse. It puts your brand at risk. Don't do it, and do watch for third-party pages that misrepresent you this way.

Reporting

Build a report that shows status, not a promised score:

• Access status per crawler: allowed, blocked, or blocked at the firewall.
• Rendering gaps found and closed.
• Wrong facts open and closed, by engine.
• Listings corrected.
• The per-engine baseline re-run, with grounded and ungrounded answers kept apart.

Never report an AI rank. Google's Search Console report for AI features shows impressions only, with no position data to check a rank against. Bing Webmaster Tools' AI Performance report is the one place that shows grounding queries, the searches the AI actually ran.

• robots.txt reviewed per bot, with training and search bots treated separately
• Each AI crawler requested a live page and got a 200, not a 403 or a challenge
• CDN bot settings checked for the September 2026 Cloudflare defaults, if you use Cloudflare
• Raw HTML contains pricing, product names and core claims without JavaScript
• No accidental noindex, nosnippet, max-snippet:0 or data-nosnippet
• Search Console generative AI inclusion control checked where visible
• Fixed buyer question set saved, with baseline answers per engine
• Fact sheet written and signed off
• Contradicting pages on your site updated or retired
• Wrong facts logged by engine, question and retrieval state
• Cited third-party domains listed and sorted
• Editable listings corrected to match the fact sheet
• Google and Bing index status checked separately
• Unsupported GEO tasks removed from the roadmap
• First quarterly report sent, with no rank and no blended score

For a longer version organised by the same three themes, use the AI SEO checklist.

Judge the quarter on what you control. Mentions and citations are outputs the models decide.

Good results

• Every retrieval crawler you intend to allow reaches a live page.
• Your key facts appear in raw HTML on every page that states them.
• Wrong facts in grounded answers trace to pages you've fixed, and the count drops on re-run.
• Your report separates engines and says which answers used live retrieval.
• Leadership understands the baseline is a sample, not a census.

Bad results

• robots.txt allows a crawler that the firewall still blocks.
• The question set changed mid-quarter, so the before and after can't be compared.
• The team spent month two on schema, llms.txt or rewriting for "AI style".
• One blended score went up or down and nobody can say why.

One result is neither good nor bad: ungrounded answers still repeating an old fact. That comes from training data, and it changes only when a model is retrained. Flat mention counts after 90 days don't prove the work failed either. If you want to rank the remaining work by how strong the evidence is, see AI SEO best practices, sorted by evidence.

The plan is front-loaded. After the first quarter, most of it becomes routine.

• Access checks: monthly, and after every deploy, redesign, CDN change or security rule update. Vendors change bot defaults without asking you.
• Rendering diff: after any front-end framework change or new page template.
• Fact sheet: quarterly, and the same day a price, plan or product name changes.
• Baseline re-run: monthly, with the same questions, logged out and memory off.
• Third-party listings: quarterly, starting with the domains engines cite most.
• Full 90-day cycle: once a year, or after a rebrand, acquisition or major site rebuild.

A monthly snapshot catches breaks. It doesn't show how answers move week to week. The page on what a free snapshot misses explains when continuous tracking is worth adding.

If my SEO is strong, can I skip days 1 to 30? No. Strong SEO helps, but it doesn't cover AI crawler access. Ahrefs found in March 2026 that 37.9% of URLs cited in AI Overviews also appeared in the top 10 results, so most citations came from elsewhere. Googlebot renders JavaScript and most AI crawlers don't, and Google rankings say nothing about your robots.txt rules for OAI-SearchBot or your firewall. The access checks take days, not weeks. See [whether strong SEO means AI visibility](/questions/strong-seo-ai-visibility).

Should schema markup or llms.txt be part of this plan? Not as a visibility lever. Google says structured data isn't required for its generative AI features, and its guidance says Search ignores llms.txt. No other engine documents reading llms.txt. Ahrefs' matched test of 1,885 pages found AI Overviews citations fell 4.6% after schema was added. Keep existing markup if it serves other purposes. It does no harm. Just don't spend this quarter adding it to get cited.

How do I know whether an AI answer used live retrieval? Look for citations or source links in the answer. If they're there, the engine searched. If they're absent, you can't be sure, and the answer may come from training data with a fixed cutoff. Retrieval is discretionary on every major platform and changes per question. That's why the baseline records retrieval state for each answer, and why grounded and ungrounded answers belong in separate columns of your report.

Can we promise leadership more AI citations by day 90? No, and nobody honestly can. The plan promises things you control: crawlers that can reach you, facts in fetchable HTML, one consistent fact sheet, corrected listings and a clear report. Whether an engine retrieves and cites your pages is its decision. Frame the quarter as removing blocks and fixing wrong facts, then report movement in the baseline as an observation, not a target hit or missed.

What if an AI platform keeps repeating a wrong fact after we fix our site? First check whether the answer was grounded. If it cites sources, open them. The wrong fact often sits on a third-party page, an old post of yours or a cached directory listing, so fix or request a correction there. If the answer is ungrounded, it comes from training data and will persist until the model is retrained. Log it, keep your own pages correct, and re-check each month.