LLM SEO Services: The 9-Step Delivery Run Sheet (2026)

LLM SEO services come down to one job: getting your pages retrieved and cited when ChatGPT search, Perplexity, Claude, Copilot and Google's AI features build an answer.

Training data is a side issue.

What moves the needle is the live retrieval step, where each engine fetches pages through its own crawler or index and decides which ones to link.

Below is the delivery run sheet I'd follow on a real engagement, in order, with the commands and settings at each step. If you want a team to run it on your site, book a free SEO strategy session first and bring your Search Console access.

Map the five retrieval paths before you scope anything

Every LLM-based search product gets its pages from somewhere, and the vendors document where.

Write this table into the kickoff doc, because every later step hangs off it.

Engine	Where pages come from	Control you own	Primary doc
ChatGPT search	OAI-SearchBot crawl plus third-party search partners	robots.txt for OAI-SearchBot, CDN allow-list for OpenAI IPs	OpenAI crawlers
Perplexity	PerplexityBot crawl, Perplexity-User fetches	robots.txt for PerplexityBot, WAF rules	Perplexity crawlers
Claude	Claude-SearchBot index, Claude-User fetches	robots.txt per subdomain	Anthropic crawlers
Google AI Overviews and AI Mode	Google's Search index (Googlebot)	Googlebot access, nosnippet, max-snippet, noindex	AI features and your website
Copilot and Bing AI summaries	Bing index and grounding	robots.txt, NOINDEX, NOARCHIVE, NOCACHE	Bing Webmaster Guidelines

Notice what isn't in that table: a separate trick for any one engine.

Google's generative AI guide says optimising for its AI search is "still SEO", and Bing says Copilot sits on the same crawling, indexing and ranking foundation as its classic results.

So an LLM SEO service is classic SEO, done for five retrieval systems at once.

Step 1: Take a read-only baseline on day one

Start with read access only. Google's hiring an SEO page tells site owners not to hand over write access to Search Console at the audit stage, and a good provider will ask for less, not more.

Then capture four numbers and date them:

Impressions in the Search Console generative AI performance report, live for all sites since 31 August 2026.
Total citations and cited pages in Bing Webmaster Tools AI Performance, which covers Copilot and Bing's AI summaries.
Referral sessions from chatgpt.com, perplexity.ai and claude.ai in your analytics.
A prompt log: 20 to 30 buyer questions, asked in each engine, with the cited domains written down.

On the analytics line, in the accounts we look at ChatGPT referrals usually arrive with chatgpt.com as the source, often with utm_source=chatgpt.com on the landing URL. That's our observation, not an OpenAI statement, so treat the filter as a starting point.

The prompt log is the one number no dashboard gives you, and it's the one clients care about most.

Step 2: Prove the search bots actually reach you

robots.txt tells you what you asked for; server logs tell you what happened.

Pull 30 days of access logs and count hits per AI user agent:

grep -Eo "OAI-SearchBot|GPTBot|ChatGPT-User|PerplexityBot|Perplexity-User|Claude-SearchBot|ClaudeBot|Claude-User|bingbot|Googlebot" access.log \
  | sort | uniq -c | sort -rn

# then check status codes for one bot
grep "OAI-SearchBot" access.log | cut -d'"' -f3 | cut -d' ' -f2 | sort | uniq -c

Zero hits from OAI-SearchBot, or a wall of 403s, usually means a firewall is blocking it before robots.txt is ever read.

OpenAI's ChatGPT search help article makes allowing its published searchbot IP addresses part of being eligible, and Perplexity's docs walk through Cloudflare and AWS WAF allow rules for its bots.

One Perplexity gotcha worth flagging: its docs say Cloudflare's Bot Fight Mode can't be skipped with WAF custom rules, so a skip rule alone may not be enough.

Verify claimed bots against the JSON IP lists each vendor publishes, because user-agent strings are easy to fake.

📞 Want this run sheet run on your site? Book a free SEO strategy session and the team will walk your logs, robots.txt and AI citations with you. → Get your free strategy session

Step 3: Split search bots from training bots in robots.txt

Each vendor separates the bot that powers search answers from the bot that collects training data, and the settings are independent.

OpenAI states you can allow OAI-SearchBot while disallowing GPTBot, and that changes take about 24 hours to register for search.

Perplexity says PerplexityBot is not used for foundation model training.

Anthropic warns that blocking Claude-SearchBot may reduce your visibility in Claude's search results, and asks you to set rules on every subdomain you want covered.

The decision on training bots belongs to the client, not the SEO.

What the SEO must stop is the common accident: a blanket AI block copied from a forum that takes out the search bots too.

Two lines confuse almost everyone:

Google-Extended. Google's common crawlers page says: "Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search." It governs Gemini training and grounding in Gemini Apps and Vertex AI, not AI Overviews.
ChatGPT-User. It fetches pages when a user asks, robots.txt "may not apply", and OpenAI says it isn't used to decide what appears in search.

Step 4: Audit preview controls page by page

Snippet controls now decide how much of a page an AI answer can use, so audit them like you'd audit canonicals.

Google's robots meta tag documentation says nosnippet also stops content being used as a direct input for AI Overviews and AI Mode, and max-snippet caps how much can be used.

On Bing, the guidelines say NOARCHIVE keeps content out of Copilot responses and NOCACHE limits Copilot to the URL, title and snippet.

Diary note for 2027: Bing has announced new AI Training and AI Grounding preferences that take effect on 1 February 2027 and replace NoArchive and NoCache for those uses, with an action deadline of 1 March 2027.

Crawl every indexable URL and export the robots meta and X-Robots-Tag values into one sheet.

Flag any money page carrying nosnippet, a tiny max-snippet, NOARCHIVE or NOCACHE, and confirm whether that was a deliberate choice.

Step 5: Make each money page stand on its own

Bing's guidelines are unusually specific here, and the advice travels well to every engine.

They ask for explicit facts and definitions, key statements that don't depend on implied context, and one primary topic per URL.

For a service business, that means a page per service with the who, where, how long and what's included stated in plain text on the page itself.

Check the text exists in the raw HTML too; Google's JavaScript SEO basics note that not every bot can run JavaScript.

Rewrite vague hero copy into direct answers to the questions in your prompt log, then put the evidence on the same page.

Step 6: Lock down entity data and naming

An LLM can only cite you confidently when it can tell which business you are.

Pick one exact brand name, one address format and one service list, and use them everywhere: site footer, Business Profile, directories, social bios.

Add Organization markup with logo, url and sameAs links to your official profiles; Google's Organization structured data guide says this helps it disambiguate your business.

Keep expectations straight, though: Google says no special schema is needed for AI features, and structured data must match the visible text.

Some agencies also ship an llms.txt file. Google's guide is explicit that Search ignores it, so treat it as optional for other systems, never as a Google tactic.

🔥 Want to run these steps yourself? Inside the AI Profit Boardroom you get step-by-step AI SEO trainings, the installable Agent OS to run the audits, and four live coaching calls a week to check your work. → Join the AI Profit Boardroom

Step 7: Publish content an LLM can't get anywhere else

Google calls it non-commodity content: first-hand, expert material that goes past common knowledge.

Its guide says this will likely influence your presence in generative AI search more than any other suggestion it makes.

For a plumber, that's a page on the three causes of low water pressure you actually see in local housing stock, with photos from real jobs.

For an accounting firm, it's a worked example of a deadline change and what clients had to do, written by the person who did it.

What doesn't work is spinning out a page for every fan-out query; Google names that as scaled content abuse.

Step 8: Earn mentions and links inside the rules

AI answers reflect what the wider web says about you, so third-party mentions matter.

Google's guide warns that seeking inauthentic mentions isn't as helpful as it looks, and its spam policies define link spam as links made mainly to manipulate rankings.

Paid placements are fine when marked with rel="sponsored" or nofollow, per the spam policies; outreach-earned editorial links and genuine digital PR are the core of the work.

Goldie Agency describes its own link building as "Manual outreach, no PBNs", which is the standard I'd hold any provider to.

Target the sites that already appear in your prompt log citations: trade bodies, review platforms, local news, niche publications.

Step 9: Report citations, not just rankings

A monthly LLM SEO report should fit on one page and answer three questions: are we reachable, are we cited, and is it sending business?

LLM SEO monthly report
1. Access: AI search bot hits + error rate (logs)
2. Google: gen AI impressions, top pages (Search Console)
3. Bing/Copilot: citations, cited pages, grounding queries
4. ChatGPT/Perplexity/Claude: prompt log, cited yes/no, who was cited instead
5. Traffic: referral sessions + conversions by AI source
6. Next month: 3 actions, owner, due date

Bing notes that a fall in clicks doesn't always mean a fall in visibility, which is why citations get their own line.

Where AI agents speed up the delivery

Agents are good at the repetitive half of this work.

Use one to parse logs into the bot table, re-run the prompt log every month, crawl preview-control headers, and draft Organization markup for review.

Keep them on read and draft permissions; a person signs off robots.txt edits, schema deploys and every outreach email.

Five gotchas that stall engagements

Docs or help centres on a separate host with their own robots.txt that nobody checked.
A CDN bot setting that challenges OAI-SearchBot or PerplexityBot while robots.txt says allow.
Blocking Google-Extended in the belief it removes you from AI Overviews; it doesn't.
nosnippet added sitewide for a legal reason, quietly cutting AI Overviews use.
Brand name spelled three ways across site, profiles and directories.

What nobody can promise you

No provider controls these answers.

OpenAI says ChatGPT search placement "is not guaranteed", Bing says GEO doesn't guarantee grounding or citations, and Google says no one can guarantee a #1 ranking.

Treat a guaranteed mention, a "special relationship" or a secret method as a reason to walk away.

What a good service can promise is the work: access fixed, pages improved, mentions earned and citations measured.

Julian Goldie's own published numbers put his agency at 1,000+ SEO and link building campaigns and 240+ clients, with a 100% Job Success Score on Upwork; he also shares AI SEO tests publicly with 425K+ YouTube subscribers.

📞 Hand the run sheet to a team Book a free SEO strategy session and get the baseline, access audit and first three fixes mapped for your site. → Book your strategy session

🎁 Two free tools before you start Grab the 200+ AI SEO Prompts for the prompt log and the free link building book for the outreach step. Learning it in-house? The AI Profit Boardroom covers each step on video. → Book a free SEO strategy session

Also On Our Network

FAQ: LLM SEO Services

What's the first deliverable in an LLM SEO engagement?

A dated baseline: Search Console generative AI impressions, Bing AI Performance citations, AI referral traffic and a prompt log of buyer questions with the domains each engine cited.

How do I know OAI-SearchBot can reach my site?

Count its hits and status codes in your server logs. If you see none, or mostly 403s, check your CDN or firewall, since OpenAI asks hosts to allow its published IP addresses.

Does nosnippet affect AI Overviews?

Yes. Google's robots meta tag documentation says nosnippet also prevents content being used as a direct input for AI Overviews and AI Mode.

Do I need llms.txt for LLM SEO?

Not for Google. Google's AI optimisation guide says Search ignores llms.txt, so it neither helps nor harms there.

How often should citations be re-checked?

Monthly works for most sites: re-run the same prompt log, compare cited domains, and log three actions with owners for the next month.

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
I've grown a YouTube channel to 400,000+ subscribers.
I wrote "Link Building Mastery".
My Udemy courses have taught over 29,000 students.

→ Get my best AI training inside the AI Profit Boardroom

Run the nine steps in order, report citations every month, and when you want a team running them for you, book your free strategy session for LLM SEO services built on the platforms' own docs.