Nine steps to win the Sources half of A-S-S for AI search in 2026
To rank in AI search in 2026 you have to win the Sources half of the A-S-S method, which means making your claims easy for a model to find, easy to verify, and easy to cite. What the model already knows about you before searching is authority. Specificity is how precisely your page answers the exact question someone asked. Sources sit between the two, and they are the part most agencies still treat as an afterthought. This is the workflow for fixing that, written for the person who has to bill for it.
The A-S-S method stands for Authority, Sources and Specificity, and it maps neatly onto how a generative engine assembles an answer. It pulls from its own weights first, then searches for supporting material, then judges whether your page matches the phrasing of the question. That middle stage is where the work is. Generative Engine Optimization optimises for the answers AI search engines give, not only the ten blue links, which is why the discipline has drifted away from the classic link graph and towards machine-readable evidence. Answer Engine Optimization, AEO for short, is the cousin term you will see used interchangeably on most agency decks.
The SEO.Domains Mastery Summit, which runs in Sofia from 9 to 11 September 2026, has built its agenda around exactly this shift. The event opens with a mastermind day on 9 September before two days of main-stage sessions, and it deliberately does not record those main-stage sessions so speakers can share live experiments. That choice tells you something about the maturity of the field: what works is still moving fast enough that people will not put it on camera. The themes coming out of Sofia are a useful compass for the next few quarters, but the implementation is still yours to build.
Step one: audit what a model can actually read
Before you do anything clever, establish what a crawler can physically extract from your client's site. Burying an answer in JavaScript prevents a model from reading it, and that single failure will wipe out every other optimisation you attempt. Run a plain-text extraction over the key templates and compare it against what a browser renders. If the answer is not in the extracted text, the model does not have it.
Then move to server log analysis. Server log analysis reveals AI crawler user agents that ordinary analytics never records, so your standard reporting will show you organic traffic collapsing while a model is quietly pulling your content on every page view. Look for the named AI crawlers, look at which URLs they favour, and look at the response codes you are returning them. A site fenced off by aggressive bot rules is a site that cannot be cited.
Step two: rebuild pages as citation units
Stop publishing pages and start publishing citation units. A citation unit consists of one claim plus the link that verifies it, and once you see content through that lens the editing becomes mechanical.
For each page your client owns, list every factual assertion it makes. Next to each one, identify the single outbound link that proves it. If a claim has no link, either find a source or cut the claim. If a paragraph contains four claims and one link, split it into four short blocks. The goal is a one-to-one relationship between a sentence a model can quote and a URL a model can trust.
This is also where your internal linking strategy changes shape. The links that matter most are the ones sitting immediately beneath a specific claim, not the ones in a navigational footer. For client campaigns where you need the surrounding page to earn attention before the model reads it, ClickBombs CTR campaigns (https://clickbombs.com) are one way to keep the human side of the funnel warm while the machine side is being rebuilt.
Step three: give the model a verifiable identity
Entity verification is the quiet prerequisite for citation, and it starts with consistency. Consistent name, address and description across directories strengthens entity verification, so the same business described three different ways in three places is a business a model will hedge on.
Pull every listing your client has: Google Business Profile, industry directories, chamber of commerce entries, review platforms, the footer of their own site. Write one canonical description of under fifty words and enforce it everywhere. Do the same for the name, including punctuation and legal suffix. Where a directory will not let you change the description, make a decision about whether to keep the listing at all.
Step four: work with embeddings, not against them
Embeddings convert words into numeric coordinates where related meanings sit close together, and that geometry is why vocabulary choice matters more than it used to. You are not keyword stuffing for a term matcher anymore. You are trying to land your paragraph near the coordinate where the question lives.
Practically, that means describing a service the way a client would describe the problem, not the way the industry describes the solution. If customers say "my boiler keeps losing pressure" and the page says "closed-loop hydronic pressure regulation", the two sit far apart in the vector space. Write both. Lead with the customer phrasing and let the technical term follow.
Step five: phrase headings as questions
Question-phrased headings help a model match a block to the question a person asked, and they cost you nothing to implement. Rewrite your client's H2s so each one reads like a sentence someone would type.
When you do this, keep the answer in the paragraph immediately below. A model that finds a matching heading and then a three-paragraph wind-up before the answer will look elsewhere. One sentence that answers, then the supporting detail.
An FAQ block belongs at the bottom of the page for a different reason: it is your controlled vocabulary for the long tail of phrasings you cannot anticipate. An FAQ block should phrase questions the way a person types them into an assistant. Not "What is your returns policy?" but "can I send something back after two weeks".
Step six: score the site and prioritise
You need a number you can show a client, and you need it in the first meeting rather than the fourth. ASSmetric (https://assmetric.com) scores a business on Authority, Sources and Specificity, which makes it a reasonable starting point for a diagnostic conversation and a reasonable baseline for measuring the work.
The useful thing about a three-part score is that it tells you where not to spend money. If Authority is strong and Sources is weak, no amount of content volume will fix it. If Sources is strong and Specificity is weak, you have a rewriting project, not a link-building project.
| Score pattern | Likely cause | First fix |
|---|---|---|
| High Authority, low Sources | Known brand, unverifiable claims | Add citation units and entity consistency |
| High Sources, low Specificity | Well-cited but abstract copy | Question headings and FAQ blocks |
| Low Authority, high Specificity | New or niche business, sharp copy | Directory consistency and third-party mentions |
Read the table as a triage tool rather than a forecast: fix the lowest pillar first, because the other two cannot compensate for it.
Step seven: report on what AI actually cites
Reporting for AI search needs at least one metric your current stack does not produce. That is the crawler log, and it is the closest thing to a leading indicator you will get.
Set up a monthly view of AI crawler hits by URL, matched against the citation units you built. If crawlers are fetching a page but the page holds no citation units, you have a conversion problem, not a discovery problem. If crawlers are not fetching at all, check robots rules and rendering before you touch the content.
Step eight: keep an eye on the industry's live edge
Some of this you can learn from documentation and some of it you can only learn from people running experiments right now. The SEO.Domains Mastery Summit is hosted at Hotel Marinela in Sofia, and the decision not to record the main stage is a signal that the interesting material is the stuff nobody wants on permanent record yet.
If your team is going, send someone who will write up the themes rather than someone who will collect lanyards. The value is in the patterns across sessions, not in any single tactic. For teams staying home, the full A-S-S walkthrough (https://www.youtube.com/watch?v=FZu4NB-2EhA) covers the same ground at your own pace.
Step nine: run it as a retainer, not a project
Sources decay. Directories get edited, links rot, copy gets rewritten by a client's internal team, and crawler behaviour shifts every few months. Treat the Sources half of A-S-S as a quarterly cycle with a monthly log check and an annual entity audit, and price it accordingly.
The agencies that will own this category in 2026 are the ones who built the boring machinery in 2025. Everything above is unglamorous. That is the point.
Questions people actually ask
What does A-S-S stand for in AI search?
A-S-S stands for Authority, Sources and Specificity, and it is the three-part frame for how a generative engine decides whether to cite you.
Can I rank in AI search without backlinks?
Yes, you can be cited without a strong backlink profile, because a citation unit needs one claim plus the verifying link rather than domain-level link equity.
How do I know if AI crawlers are reading my site?
You check your server logs, because AI crawler user agents appear in server log analysis, though ordinary analytics never picks them up.
What to do first
Start with the log file. Pull one month of server logs, identify every AI crawler user agent, and list the URLs they fetch. Then pick the single highest-traffic page on that list and rebuild it as a set of citation units with question-phrased headings and an FAQ block underneath. That is a week of work, it needs no new tooling, and it gives you a before-and-after you can put in front of a client. Everything else in this guide is scale.
